Does Speed-Normalized Double-Support Reflect Gait Stability in Parkinson’s Disease? A Model-Based Analysis

Abstract

Background: Double-support percentage (DS%) is often interpreted as a proxy for dynamic gait stability, yet its biomechanical meaning is confounded by its strong inverse coupling with walking speed. This distinction is critical in Parkinson’s disease (PD), where bradykinetic gait inherently prolongs DS%. To isolate speed-independent stability demands, we introduced a model-based Stability Reserve Index (SRI), representing the deviation between predicted and observed double support after normalizing for velocity and anthropometrics. Methods: Using an open-access dataset of 63 individuals with PD (ON medication; Hoehn & Yahr 1–3) and 63 matched controls, step-based DS% was modeled using ANCOVA, incorporating centered walking speed, group, their interaction, and covariates. Predicted DS% at the sample’s grand mean speed was subtracted from observed DS% to derive the SRI, indexing whether double support exceeded expectations for a given biomechanical operating point. Results: PD participants walked slower than controls (p < 0.001), but once velocity was accounted for, DS% no longer differed between groups (p = 0.795–0.880), and the DS%–speed coupling remained intact (interaction p = 0.387). Speed-normalized predicted DS% (p = 0.159) and the SRI (p = 0.989) were likewise similar across groups. Within PD, SRI did not correspond to UPDRS-III or Hoehn & Yahr stage (ρ = 0.129–0.223, p > 0.05). Conclusions: These findings indicate that double-support behavior in mild-to-moderate PD is largely velocity-driven rather than reflecting altered dynamic stabilization strategies. While conceptually grounded in stability reserve theory, the SRI showed limited discriminatory value under ON-medication walking, suggesting that more sensitive multidimensional metrics—integrating CoM dynamics, variability, and step-to-step control—may be required to capture early instability in PD.

1. Introduction

Locomotor stability during walking relies on the continuous regulation of segmental dynamics, foot placement, and step-to-step adjustments to maintain the center of mass (CoM) within a safe margin relative to the base of support [1]. Among the spatial–temporal gait parameters, double-support time (DS%) is a critical marker of dynamic stability, reflecting the period in which both feet are in contact with the ground and the CoM is mechanically better controlled [2,3]. Older adults and individuals with neurological disorders often prolong DS% as a compensatory strategy to enhance stability during steady-state walking [4,5,6]. However, DS% is also highly dependent on walking speed—slower gait naturally increases DS%—which complicates its direct interpretation as an indicator of impaired stability [7,8].

In Parkinson’s disease (PD), gait disturbances such as bradykinesia, impaired scaling of step amplitude, and reduced automaticity of locomotor control contribute to altered spatiotemporal organization of gait [9,10]. Individuals with PD frequently exhibit slower speed, shorter steps, and a greater proportion of double support even in the ON-medication state, suggesting that DS% may index compensatory behavior or underlying instability [11]. Yet, because PD is also associated with reduced walking velocity, distinguishing whether elevated DS% reflects true instability or simply slower gait remains a longstanding challenge in clinical gait analysis [12].

Recent biomechanical frameworks emphasize the importance of speed-normalized gait metrics to separate pathological gait changes from speed-related adaptations [13]. Modeling approaches that account for the typical relationship between DS% and walking speed may allow quantification of “excess” double support relative to what would be expected for a given speed. This concept aligns with the idea of a “stability reserve”, wherein gait stability emerges from the difference between the stability that an individual should require at a given walking speed and the stability that is actually utilized [2]. Despite the conceptual relevance, few studies have attempted to derive individualized, speed-normalized stability markers in PD. Previous work has typically addressed the confounding influence of gait speed by either matching or imposing walking speeds between patients and controls, or by normalizing gait parameters such as gait speed to body height or leg length [14,15]. Other studies have modeled the effect of speed on spatiotemporal parameters, including double support, using multi-speed protocols and regression or ANCOVA-based approaches, thereby highlighting that part of the observed group differences can be attributed to gait velocity rather than pathology per itself [16]. However, these approaches generally provide group-level corrections and have not yielded an individualized stability index that quantifies how much double support deviates from the value predicted from a person’s walking speed and anthropometric characteristics.

To address this gap, we propose a Stability Reserve Index (SRI), which is a metric derived by comparing model-predicted and actual double-support percentages (DS%) at the population’s mean walking speed. By quantifying deviations from expected DS% when speed-related effects are controlled, the SRI provides a speed-normalized indicator of stability utilization. Using a well-characterized open dataset of older adults with and without PD, we examined (1) whether the DS%–speed relationship differs between groups, (2) whether individuals with PD exhibit reduced stability reserve, and (3) whether SRI corresponds to established clinical markers of disease severity. We hypothesized that PD participants would show greater reliance on double support than expected for their walking speed, and that lower SRI values would be associated with worse motor impairment.

2. Materials and Methods

2.1. Study Design

This cross-sectional analytical study utilized de-identified gait and clinical data from an open-access dataset originally collected at the Human Movement Research Laboratory (MOVI-LAB), São Paulo State University (UNESP), Brazil [17]. The dataset includes synchronized spatial–temporal gait parameters and clinical assessments obtained during standardized laboratory walking trials. The present study performed an independent secondary analysis focusing on double-support–based stability metrics and the SRI.

2.2. Participants

The dataset comprised 126 community-dwelling older adults, including 63 individuals with PD and 63 age-matched healthy controls [17]. In total, the sample included 74 females and 52 males, with ages ranging from 52 to 84 years, and the two groups were matched at the group level for age, body mass, and body height (see

Table 1. Participant characteristics by group.

Variable	Control (n = 63)	PD (n = 63)	Between-Group Difference (Control − PD), 95% CI	p-Value
AGE, YEARS	68.32 ± 6.28	68.79 ± 7.51	−0.48 (−2.92 to 1.96)	0.700
BODY MASS (KG)	72.05 ± 13.21	69.41 ± 13.75	2.64 (−2.11 to 7.40)	0.273
BODY HEIGHT (M)	1.598 ± 0.081	1.62 ± 0.09	−0.02 (−0.05 to 0.01)	0.131
BMI (KG/M2)	28.20 ± 4.66	26.37 ± 4.66	1.82 (0.18 to 3.47)	0.030 *
MMSE (SCORE)	27.95 ± 2.02	27.49 ± 2.00	0.46 (−0.25 to 1.17)	0.201
SEX, N (%)
FEMALE	45 (71.4%)	29 (46.0%)	χ2(1) = 8.38	0.004 **
MALE	18 (28.6%)	34 (54.0%)

Note: ** indicates a significant difference at p-value < 0.01, * indicates a significant difference at p-value < 0.05.

for detailed characteristics). All participants had preserved global cognition (Mini-Mental State Examination score ≥ 24, adjusted for education) and were able to walk independently without an assistive device. All assessments were originally conducted under standardized laboratory procedures. Participants with PD were evaluated in their ON-medication state, approximately 1 h after dopaminergic intake, and were classified within Hoehn & Yahr stages 1–3, with independent ambulation [18]. Controls had no neurological, orthopedic, or cognitive impairments affecting gait. Only self-selected comfortable-speed walking trials were included in the present analysis. All participants had previously provided informed consent during the original data collection, and the dataset was publicly released in anonymized form, permitting secondary analysis without additional ethical approval [17].

2.3. Dataset Description

The dataset contains demographic information (age, sex, height, body mass, BMI), cognitive scores (MMSE), and clinical severity indices for PD (UPDRS-III, Hoehn & Yahr) [19,20,21]. Gait data were collected using a GAITRite^® walkway system (GAITRite Gold, CIR Systems, Havertown, PA, USA) (8.5 x 1.5 m length x width; 5.74 m active area; 100 Hz), integrated with a 10 Vicon Motion Systems® cameras (100 Hz) [17]. Participants performed level-ground walking at a comfortable speed, with trials initiated several steps before and after the sensing area to minimize acceleration or deceleration effects. Each participant completed three valid trials, and the dataset provides average spatial–temporal parameters automatically computed by the Matlab software (version 7.10, Mathworks) [17]. Only these summary values were used in the present analysis. No alterations were made to the raw dataset aside from variable cleaning and generating derived metrics (e.g., speed-normalized predicted double-support and SRI).

2.4. Clinical Assessments

Motor symptom severity in PD was evaluated using the Unified Parkinson’s Disease Rating Scale, Part III (UPDRS-III), whereas disease stage was determined by the Hoehn & Yahr (H&Y) scale (stages 1–3 in this dataset) [19,20,21]. Cognitive status for all participants was assessed using the Mini-Mental State Examination (MMSE). UPDRS-III and H&Y scores were used to examine clinical correlates of gait SRI, while demographic and anthropometric variables were used descriptively and as covariates where appropriate.

2.5. Gait Data Acquisition

Walking trials were conducted in a controlled laboratory environment following the standardized procedures outlined by Penedo et al. (2024) [17]. Participants walked barefoot along the GAITRite walkway at a natural, comfortable pace. Passive reflective markers placed according to the Plug-in-Gait protocol enabled synchronized collection of marker trajectories and footfall data through the combined GAITRite–Vicon system. Trials were considered valid if participants walked continuously without hesitation or targeting the sensing area. Mean values across valid trials were used for all analyses. All PD assessments occurred during ON-medication conditions.

2.6. Data Processing

All data processing procedures were performed using IBM SPSS software version 26.0 (IBM SPSS Statistics, SPSS Inc., Chicago, IL, USA). Extracted variables included step- and stride-based walking velocity, support-phase durations (double and single support), demographic characteristics (age, sex, height, body mass, BMI), and PD-specific clinical measures (UPDRS-III and Hoehn & Yahr stage). Step-based double-support duration was expressed as the percentage of the gait cycle (step DS%). Walking velocity (v, m/s) was centered around the grand mean of the entire sample to obtain a centered-velocity term $\mathrm{v}\_\mathrm{c}=\mathrm{v}-\overline{\mathrm{v}}$ which allowed us to interpret model effects at the typical walking speed of all participants.

For analytic modeling, groups were coded as PD = 0 and Control = 1. Walking velocity was centered around the grand mean of the entire sample, allowing the velocity × group interaction to be interpreted at the typical walking speed of all participants while minimizing multicollinearity. A general linear model (ANCOVA) predicting step-based DS% included centered velocity, group, the velocity × group interaction, and covariates (age, height, BMI). The full model was specified as follows:

step DS% = β₀ + β₁(v_c) + β₂(G) + β₃(v_c × G) + β₄(age) + β₅(height) + β₆(BMI) + ε,

where v_c denotes the centered walking velocity, G is group membership (0 = Parkinson’s disease, 1 = control), each β coefficient represents a fixed effect, and ε is the residual error term.

Model fixed-effect coefficients were then used to estimate each participant’s predicted double-support value at the sample’s grand mean walking speed (i.e., where the centered velocity equals zero). Model fit indices, including the adjusted R² and the standard error of the estimate (SEE), were computed to evaluate the accuracy of the ANCOVA model underlying the predicted DS% values used in the SRI calculation. This value represents the speed-normalized expected double-support for each participant based on group membership and anthropometric covariates. In practical terms, this procedure first removes between-participant differences in raw walking velocity by centering individual speeds around the grand mean. Second, the ANCOVA model quantifies how step-based double support changes as a function of this centered velocity while simultaneously adjusting for group and anthropometric covariates. Third, by evaluating the fitted model at v_c = 0, we obtain the double-support value that each participant would be expected to exhibit if they were walking at the common reference (grand mean) speed. Finally, computing SRI as the difference between the observed and predicted values yields an individualized index of double support that is normalized for walking speed and basic anthropometrics. The SRI was subsequently computed as follows:

$$\mathrm{SRI}(\%)={\mathrm{DS}}_{\mathrm{pred}\_\mathrm{ref}}-\mathrm{step}\mathrm{DS}\%$$

where DS_{pred_ref} denotes the predicted step-based double-support percentage at the reference (grand mean) walking speed obtained from the ANCOVA model, and step DS% is the participant’s observed step-based double-support percentage. Because both DS_pred_ref and step DS% are expressed as double-support duration as a percentage of the gait cycle, SRI is also expressed in percentage points (% of the gait cycle). Thus, more negative values indicate reduced stability reserve, that is, a greater reliance on double support than predicted for the reference speed.

For clinical analyses restricted to the PD group, associations between SRI and UPDRS-III or Hoehn & Yahr stage were examined using Spearman’s rho, followed by multiple regression to evaluate the combined contribution of clinical and anthropometric factors. Assumptions of normality, linearity, and homoscedasticity were evaluated using Shapiro–Wilk tests, and Levene’s tests; nonparametric tests were applied where assumptions were violated.

2.7. Ethical Considerations

This study involved secondary analysis of fully de-identified, publicly accessible human data and therefore required no additional ethical approval. The original data collection was approved by the São Paulo State University (UNESP) Research Ethics Committee (protocol CAAE#78660517.2.0000.5398) [17], and all participants provided written informed consent. All procedures adhered to the Declaration of Helsinki and current standards for ethical secondary use of open-access data.

2.8. Statistical Analysis

All statistical analyses were conducted using SPSS software version 30.0 (IBM SPSS Statistics, SPSS Inc., Chicago, IL, USA), and all figures were produced in GraphPad Prism 10 for macOS (Version 10.3.1). Between-group differences in demographic variables and baseline gait parameters were assessed using independent-samples t-tests, with sex distribution compared using chi-square tests. Means, standard deviations, and 95% confidence intervals are reported. Group differences in spatial–temporal gait variables, as well as in speed-normalized predicted double-support and SRI, were evaluated using independent-samples t-tests. The ANCOVA model described in Section 2.6 was used to examine the effects of centered walking velocity, group, and their interaction on DS%. SRI values derived from this model were subsequently used to evaluate clinical associations within the PD group. Spearman’s rho was used to assess correlations between SRI, UPDRS-III, and Hoehn & Yahr stage, followed by multiple regression analyses to identify the strongest predictors of SRI. All analyses used two-tailed tests with α = 0.05.

3. Results

3.1. Participant Characteristics

Descriptive characteristics of the sample are summarized in Table 1. The PD and control groups were comparable in age, body mass, body height, and global cognitive status as assessed by the MMSE (all p ≥ 0.131). BMI was slightly higher in the control group than in the PD group (mean difference 1.82 kg/m², 95% CI 0.18–3.47, p = 0.030). The sex distribution differed significantly between groups, with a greater proportion of women in the control group and more men in the PD group (71.4% vs. 46.0% female; χ²(1) = 8.38, p = 0.004).

3.2. Spatiotemporal Gait Parameters and Speed-Normalized Stability Indices

Group differences in gait parameters and stability indices are summarized in

Table 2. Spatiotemporal gait parameters and stability indices by group.

Variable	Control (n = 63)	PD (n = 63)	Between-Group Difference (Control − PD), 95% CI	p-Value
Step velocity (m/s)	1.21 ± 0.17	1.07 ± 0.18	0.14 (0.08 to 0.20)	<0.001 **
Stride velocity (m/s)	1.21 ± 0.18	1.07 ± 0.18	0.14 (0.08 to 0.20)	<0.001 **
Step DS% (% gait cycle)	36.78 ± 9.92	37.28 ± 11.30	−0.49 (−4.24 to 3.26)	0.795
Stride DS% (% gait cycle)	35.89 ± 9.95	36.61 ± 10.63	0.28 (−3.35 to 3.91)	0.880
Speed-normalized predicted double-support (%)	36.42 ± 9.73	36.89 ± 11.97	−0.47 (−1.12 to 0.19)	0.159
Stability Reserve Index (SRI) (%)	−3.36 ± 9.64	−3.38 ± 11.24	0.03 (−3.67 to 3.72)	0.989

Note: ** indicates a significant difference at p-value < 0.01.

. Self-selected walking velocity was significantly lower in the PD group than in controls for both step-based and stride-based estimates (mean difference = 0.14 m/s, 95% CI 0.08–0.20, p < 0.001). In contrast, mean double-support time did not differ significantly between groups, whether expressed on a step or stride basis (all p ≥ 0.79). The speed-normalized predicted double-support time and the derived SRI were also comparable between groups (mean differences −0.47% and 0.03%, respectively; all p ≥ 0.16) (Figure 1), indicating that, despite their slower walking speed, individuals with PD maintained a double-support profile that was broadly similar to that of healthy controls when normalized for speed and anthropometrics. The relationship between walking speed and step-based DS% for each group is illustrated in Figure 2.

3.3. ANCOVA Model Performance

The ANCOVA model predicting step-based double support from centered walking speed, group, their interaction, and anthropometric covariates did not reach statistical significance (F(6,119) = 0.77, p = 0.59). The model explained 3.8% of the variance in step-based double-support percentage (R² = 0.04, adjusted R² = −0.01), with a standard error of the estimate of 10.65 percentage points (% of the gait cycle). These values indicate that, within this relatively homogeneous cohort, step-based double support varied only weakly with walking speed, group, and basic anthropometric factors, which is consistent with the descriptive finding that DS% was broadly similar across groups and walking speeds.

3.4. Relationship Between Walking Velocity and Double-Support

To examine whether double-support time scaled differently with walking speed between groups, we fitted an ANCOVA model with step DS% as the dependent variable, group (PD vs. control) as the fixed factor, and centered step velocity, age, body height, and BMI as covariates. Within this model, neither the main effect of centered step velocity nor the group × centered step velocity interaction reached statistical significance (β = −4.94, t = −0.64, p = 0.522 for centered step velocity; β = 9.51, t = 0.87, p = 0.387 for the interaction), and the adjusted main effect of group was also non-significant (β = −0.13, p = 0.952;

Table 3. Parameter estimates from the ANCOVA model predicting step double-support percentage (DS%).

Parameter	β	Std. Error	t	p-Value	95% CI for β
Intercept	10.553	23.401	0.451	0.653	−35.78 to 56.89
Group (Control = 1)	−0.126	2.107	−0.060	0.952	−4.30 to 4.05
Centered step velocity	−4.936	7.681	−0.643	0.522	−20.15 to 10.27
Group × Centered step velocity interaction	9.507	10.959	0.868	0.387	−12.19 to 31.21
Age (years)	0.255	0.141	1.806	0.073	−0.03 to 0.54
Body height (m)	6.144	11.606	0.529	0.598	−16.84 to 29.13
BMI (kg/m2)	−0.044	0.210	−0.208	0.836	−0.46 to 0.37

Note: Dependent variable: Step DS%. Model includes group, centered step velocity, group × centered step velocity interaction, age, body height, and BMI as predictors. CI = confidence interval.

), indicating that, indicating that, over the relatively narrow range of self-selected walking speeds observed in this cohort, DS% did not systematically covary with speed nor differ in its speed-dependence between PD and control participants after adjustment for anthropometrics. Age showed a trend-level association with higher double-support time (β = 0.26, t = 1.81, p = 0.073), whereas body height and BMI were not significant predictors (both p > 0.59). These estimates were nevertheless used to derive the speed-normalized predicted double-support values and the SRI for each participant.

3.5. Clinical Correlates of the Stability Reserve Index

In the PD group, Spearman correlations showed only weak and non-significant associations between SRI and conventional clinical severity measures as summarized in

Table 4. Correlations and regression analysis examining clinical predictors of the Stability Reserve Index (PD group only, n = 63).

A. Spearman correlations
Variables			ρ	p-Value
SRI vs. UPDRS-III			0.129	0.313
SRI vs. Hoehn & Yahr stage			0.223	0.079
UPDRS-III vs. Hoehn & Yahr			0.545	<0.001
B. Multiple regression predicting SRI
Predictor	β	Std. Error	t	p-Value	95% CI for β
Intercept	−17.470	36.145	−0.483	0.631	−89.85 to 54.91
UPDRS-III	0.061	0.160	0.380	0.705	−0.26 to 0.38
Hoehn & Yahr stage	4.607	3.725	1.237	0.221	−2.85 to 12.07
Age (years)	0.014	0.200	0.009	0.946	−0.39 to 0.41
Body height (m)	3.130	16.829	0.186	0.853	−30.57 to 36.83
BMI (kg/m2)	−0.013	0.336	−0.040	0.968	−0.69 to 0.66

. SRI was not significantly related to UPDRS-III motor scores (ρ = 0.13, p = 0.313) (Figure 3) and showed only a small, non-significant trend toward higher values with more advanced Hoehn & Yahr stages (ρ = 0.22, p = 0.079). As expected, UPDRS-III and Hoehn & Yahr stage were moderately correlated with each other (ρ = 0.55, p < 0.001), confirming the internal consistency of the clinical severity metrics in this dataset.

To further explore whether SRI could be explained by clinical severity and basic anthropometrics, we ran a multiple linear regression model with SRI as the dependent variable and UPDRS-III, Hoehn & Yahr stage, age, body height, and BMI as predictors. The overall model was not significant (F(5,57) = 0.68, p = 0.639), accounting for only 5.6% of the variance in SRI (R² = 0.056, adjusted R² = −0.026). None of the individual predictors reached statistical significance (all p > 0.20). These findings suggest that, within this sample, the Stability Reserve Index captured aspects of dynamic gait stability that were largely independent of conventional clinical severity scales and simple anthropometric measures.

4. Discussion

This study examined how double-support behavior varies as a function of walking speed and disease status, and introduced a speed-normalized metric—the SRI—to quantify deviations between expected and actual double-support utilization. Three key findings emerged. First, although individuals with PD walked significantly slower than healthy older adults, the group difference in DS% largely disappeared once the effect of speed was accounted for. Second, the relationship between DS% and speed did not differ significantly between groups, suggesting that PD participants modulate double support according to velocity in a manner similar to healthy controls. Third, the Stability Reserve Index showed no significant group difference and was not correlated with UPDRS-III or Hoehn & Yahr stage, indicating that stability reserve, as defined in this study, may not be strongly influenced by mild-to-moderate PD severity. Contrary to our initial hypothesis, the SRI did not differentiate PD from controls and showed no clear relationship with clinical severity.

Double support is a well-established stabilizing strategy during gait, with longer DS% indicating greater reliance on bilateral limb support to manage center-of-mass movement [22,23]. However, DS% is inherently and strongly speed-dependent, as slower walking naturally increases the proportion of the gait cycle spent in double support [24,25]. The present findings reinforce this principle: PD participants showed longer DS% primarily because they walked more slowly, not necessarily because they utilized atypical stability strategies. Once velocity was centered and included as a covariate, the difference between groups was substantially reduced.

The absence of a significant group × speed interaction suggests that the mechanical coupling between walking speed and DS% is preserved in PD, at least in mild-to-moderate stages. This aligns with the view that certain fundamental spatiotemporal relationships, such as speed–cadence or speed–step length coupling, remain biomechanically intact despite deficits in automaticity and amplitude regulation [26,27,28,29]. Thus, while PD impairs the scaling of movement magnitude, it may not fundamentally alter the way gait stability scales with speed.

The SRI was designed to separate speed-related contributions to DS% from disease-related contributions. Conceptually, a lower (more negative) SRI indicates excess double support relative to what would be expected at the sample’s typical (grand mean) walking speed, potentially reflecting impaired balance, increased cautiousness, or altered sensory–motor integration [30,31,32]. However, SRI did not differ between groups and showed weak associations with clinical severity. There are several potential explanations for this. First, participants with PD were assessed in the ON-medication state, which is known to partially normalize spatial–temporal gait parameters [33,34]. Dopaminergic treatment improves stride length, speed, and rhythmicity, potentially reducing between-group differences in stability-related parameters [34,35]. Second, most individuals in this dataset were within Hoehn & Yahr stages 1–3, where gait instability may not yet be prominent. Reactive balance impairment and postural instability typically emerge in later disease stages [36,37], which were not represented here. Third, DS% may not capture the most sensitive aspects of gait instability in PD. Evidence increasingly suggests that variability-based metrics, phase-dependent trunk accelerations, and step-to-step regulation strategies may provide more sensitive markers of impaired dynamic stability than mean DS% alone [5,38,39]. Given this, SRI—while theoretically appealing—may require complementary measures or nonlinear modeling approaches to better detect subtle balance impairments.

An important methodological consideration is that the present implementation of SRI was based exclusively on mean spatiotemporal parameters. Numerous studies have shown that stride-to-stride variability in step time, step length, and double support is more sensitive to early gait instability in Parkinson’s disease than central tendency measures alone. Because the open-access dataset did not provide variability metrics or fluctuation-based indices of double support, our analysis could not incorporate this dimension of gait control. It is therefore plausible that the limited discriminatory power and weak clinical associations of SRI observed here partly reflect the fact that mean double-support duration remained relatively stable, whereas potentially more informative variability in double support and related parameters could not be quantified.

Taken together, our findings suggest that, at least in relatively high-functioning individuals with mild to moderate PD who are assessed in the ON-medication state during comfortable straight-ahead walking, the SRI has limited utility as a stand-alone cross-sectional discriminator or surrogate marker of clinical severity. Rather than a diagnostic index, we therefore view SRI primarily as a methodological tool for decomposing double-support duration into its speed-related and residual components. This framework may become more informative in contexts where gait instability is more pronounced or systematically stressed, such as OFF-medication assessments, individuals with moderate to advanced PD or a history of recurrent falls, or experimental paradigms that manipulate task demands through dual-task walking, turning, obstacle negotiation, or external perturbations. In such settings, residual deviations of double support from speed-predicted values may better capture maladaptive stability strategies or emerging postural instability than in the relatively stable walking conditions examined here.

Previous studies consistently report greater double-support time in PD, even during ON-medication walking [11,40]. However, very few studies statistically adjust for walking speed, and those that do often find that the group effect is attenuated or disappears [41,42]. The present findings reinforce that slower walking is a dominant driver of increased DS% in PD, rather than an isolated disease-specific abnormality. Efforts to normalize gait metrics for speed have gained traction in biomechanics research. Methods such as dimensionless scaling, speed-matched comparisons, and regression-based predictions have been used to distinguish speed effects from pathological effects [13,43]. The SRI approach used here aligns with these methods, offering an individualized way to compare expected and observed stability demands. However, our findings suggest that DS%-based normalization alone may not fully capture disease-specific instability in earlier PD.

Clinically, the results emphasize that DS% should be interpreted with caution in PD without considering walking speed. Observed increases in DS% may reflect a benign slowing strategy rather than instability per se. The SRI, although not group-discriminative in this dataset, may still offer utility in other contexts—such as identifying excessive cautiousness, monitoring fall risk trajectories, or evaluating responses to gait rehabilitation—especially in cohorts with more advanced PD or OFF-medication status. Scientifically, the findings highlight the need for multidimensional stability markers that integrate CoM dynamics, sensorimotor processing, and step-to-step control rather than relying solely on mean temporal parameters. Future work may benefit from combining SRI-like predictions with trunk acceleration signals, nonlinear variability measures, or state-space stability metrics (e.g., Lyapunov exponents) [38,44].

A notable strength of this study is the use of a large, publicly available, well-standardized dataset, which enhances transparency and reproducibility. The modeling approach also represents a conceptually grounded attempt to isolate pathological stability changes from the confounding influence of walking speed. However, this study has several limitations that should be acknowledged. First, the analyses were based on a secondary examination of a single, open-access dataset with a modest sample size of community-dwelling older adults. Second, the PD sample consisted of individuals with mild to moderate disease severity who were assessed in the ON-medication state while walking at their comfortable self-selected speed on a level walkway. As such, the present results may underestimate potential differences in SRI that could emerge in individuals with more advanced PD, in OFF-medication conditions, or during more challenging tasks such as dual-task walking, turning, or obstacle negotiation. Moreover, the present analysis relied solely on mean spatiotemporal parameters. Gait variability is widely recognized as a sensitive marker of instability and fall risk in PD, yet stride-to-stride variability measures of double support and related parameters were not available in the source dataset. Future studies should therefore combine the SRI framework with detailed variability-based metrics and fluctuation analyses of spatiotemporal gait parameters to better capture subtle instability that may not be reflected in mean values alone. Moreover, future studies should also examine the behavior and clinical utility of SRI across a broader range of disease stages, including people with moderate to advanced PD and a history of falls, and in both ON- and OFF-medication states. Longitudinal designs are needed to determine whether SRI can predict future gait instability or fall risk beyond traditional spatiotemporal parameters. In addition, extending the SRI framework to other neurological disorders that compromise gait stability—such as stroke, cerebellar ataxia, peripheral neuropathy, or atypical parkinsonian syndromes—may help clarify whether speed-normalized deviations in double support provide a useful, generalizable marker of impaired dynamic stability across conditions.

5. Conclusions

The present study demonstrated that DS% in PD is strongly influenced by walking speed and that the speed–DS% relationship remains biomechanically intact in mild-to-moderate disease. After normalization for speed, individuals with PD did not differ from controls in predicted or observed stability demands, and the Stability Reserve Index showed limited clinical association. These findings highlight the central role of walking speed in interpreting DS%-based stability measures and underscore the need for more sensitive, multidimensional stability metrics for detecting early gait instability in PD.