Association of tibial acceleration during walking to pain and impact loading in adults with knee osteoarthritis

Abstract

Background

Higher impact loading during walking is implicated in the pathogenesis of knee osteoarthritis. Accelerometry enables the measurement of peak tibial acceleration outside the laboratory. We characterized the relations of peak tibial acceleration to knee pain and impact loading during walking in adults with knee osteoarthritis.

Methods

Adults with knee osteoarthritis reported knee pain then walked at a self-selected speed on an instrumented treadmill for 3 minutes with an ankle-worn inertial measurement unit. Ground reaction forces and tibial acceleration data were sampled for 1 minute. Vertical impact peaks, and average and peak instantaneous load rates were determined and averaged across 10 steps. Peak tibial acceleration was extracted for all steps and averaged. Pearson’s correlations and multiple linear regression analyses assessed the relation of peak tibial acceleration to pain and impact loading metrics, independently and after controlling for gait speed and pain.

Findings

Higher peak tibial acceleration was associated with worse knee pain (r=0.39; p=0.01), and higher vertical average (r=0.40; p=0.01) and instantaneous (r=0.46; p=0.004) load rates. After adjusting for gait speed and pain, peak tibial acceleration was a significant predictor of vertical average (R²=0.33; p=0.003) and instantaneous (R²=0.28; p=0.02) load rates, but not strongly associated with vertical impact peak.

Interpretations

Peak tibial acceleration during walking is associated with knee pain and vertical load rates in those with knee osteoarthritis. Clinicians can easily access measures of peak tibial acceleration with wearable sensors equipped with accelerometers. Future work should determine the feasibility of improving patient outcomes by using peak tibial acceleration to inform clinical management.

1.0. Introduction

Knee osteoarthritis (OA) is a leading cause of disability in the United States,^1,2 yet no disease modifying treatment exists. Pain, the most common symptom of knee OA, is associated with functional limitations³ and the primary reason patients seek clinical care.⁴ The mechanisms of OA-related pain are complex and likely contribute to the challenges in developing interventions that relieve pain. Because knee pain is often evoked by aberrant joint loading,^5,6 the identification and modification of biomechanical risk factors associated with knee OA are a salient point of interest.

Amongst biomechanical risk factors, the external knee adduction moment (KAM) is widely studied because it is considered a surrogate measure of medial tibiofemoral joint loading and associated with the progression of knee OA.^7,8 Various gait modification strategies for knee OA have effectively decreased KAM in laboratory environments,^9–15 yet implementation in clinical settings remain elusive. The lack of clinical implementation can be attributed to a number of factors. These include the need for expensive equipment, long data processing time, and space associated with the motion capture environment required to deliver the intervention and evaluate joint kinetics and kinematics.

Impact loading, defined as the loading that occurs immediately after the foot contacts the ground, is another metric that may contribute to knee pain.¹⁶ Impact loading is quantified in terms of the magnitude and loading rate of the initial vertical ground reaction force peak (GRF). Importantly, the peak GRF has been noted to be a primary contributor to KAM.^17,18 Higher impact loading is associated with cartilage damage and the prevalence of knee OA.^19–21 Measurement of these force variables, similar to KAM, typically involve force plates or an instrumented treadmill which represent a potential barrier to clinical implementation.

Wearable inertial measurement units (IMUs) offer a more affordable and simple way to estimate impact loading via tibial accelerometry. Based on early work from Hennig and Lafortune²² and studies in runners, it is well understood that peak tibial acceleration (PTA; the maximum acceleration along the long axis of the tibia) occurs during the early stance phase of the gait cycle (Figure 1), and is associated with vertical load rates,^23–26 and running-related injuries.^27,28 As a result, gait modification interventions aimed to reduce PTA have been used to reduce lower extremity loading in runners.^29,30 This relationship has yet to be determined in the context of knee OA but could have practical implications for the development of clinically accessible interventions that manage symptoms. Therefore, the relation of PTA during walking to pain and impact loading in adults with knee OA needs to be determined.

Figure 1.

Representative data of vertical ground reaction force (GRF; black) and tibial acceleration (TA; grey)

The primary purpose of this study was to determine the relation of PTA to 1) knee pain severity, 2) vertical impact peaks, and 3) vertical average and instantaneous load rates while walking in adults with knee OA. After adjusting for gait speed and pain which are known to influence loading, we also aimed to determine whether PTA predicted impact loading metrics.

2.0. Methods

2.1. Participants

This study involved a convenience sample of participants from a randomized feasibility study.³¹ Participants were recruited from social media advertisements in the greater metropolitan area of Boston, MA. Those who met the National Institute for Health and Care Excellence (NICE) clinical definition of knee OA were enrolled. The NICE guidelines include individuals who are at least 45 years old; have had knee pain for at least 3 months; and have knee stiffness after periods of inactivity (e.g., sleeping) that lasts less than 30 minutes.³² In addition to these criteria, individuals were asked to rate the pain in each knee when walking from 0–10 (0 = no pain; 10 = worst pain imaginable) and included if they reported ≥ 3 out of 10 in either knee. Individuals were excluded if they walked with an assistive device, had a history of knee replacement or lower extremity surgery within the last year, or had any condition (e.g., neurological) that affected lower extremity function. Those who met the eligibility criteria provided verbal and written consent prior to participating. The study protocol was approved by the Institutional Review Board at Northeastern University (#18-05-11).

2.2. Procedures

Knee pain severity was assessed with the question, “How bad has the pain been in your knee, on average, in the past week?” and scored on a Visual Analog Scale (VAS). The VAS is a valid and reliable measure of pain intensity and often used in clinical settings and research of knee OA.^33–35 Additionally, clinical use of the single-item VAS is relatively simple compared to longer questionnaires that require specific scoring methods. Participants responded for each knee separately by marking a 10 cm line where 0 cm represents “no pain” and 10 cm represents “worst possible pain”. The index knee was identified as the knee with the greatest amount of pain. When participants indicated equal pain severity for both knees, the index knee was identified by asking “Which knee causes more difficulty while walking?” and they were required to select right or left.

Participants wore standardized footwear (Gel-Excite 6, ASICS Corp., Kobe, Japan) and height and mass were measured using a stadiometer and digital scale, respectively. To determine gait speed, participants completed five, overground 10-m walk tests at their self-selected, comfortable speed. The speed of each trial was measured with timing gates (TCi system, Brower Timing Systems, LLC, Draper, UT, USA) from the middle 6-m to avoid initial acceleration and terminal deceleration.

An IMU (IMU Blue Thunder; IMeasureU Ltd., Auckland, New Zealand; dimensions: 40 mm x 30 mm x 15 mm, mass: 12 gm, acceleration range: ± 16 g) was firmly secured with a manufacturer-provided Velcro strap, immediately proximal and anterior to the medial malleolus of the index limb so that the y-axis of the sensor was aligned with the long axis of the tibia (Figure 2). Participants walked on an instrumented, split-belt treadmill (Bertec, Inc., Columbus, OH, USA) for a 5-minute warm-up and acclimation period. The treadmill speed was set to 85% of the participants average gait speed from the overground 10-m walk tests.³⁶ The treadmill speed was lowered by an additional 10–23% for 6 participants who felt that the initial speed was too fast and not sustainable. Once a comfortable treadmill speed was set, participants walked for 3 minutes. Triaxial IMU data were sampled at 1000 Hz and GRFs were sampled at 1200 Hz during the last minute of the 3-minute walk.

Figure 2.

Placement of inertial measurement unit immediately proximal and anterior to the medial malleolus, with (A) and without (B) the strap provided by the manufacturer

2.3. Data Processing

A custom-written MATLAB program (MathWorks Inc., Natick, MA, USA) was used to process raw accelerations from the IMU. PTA was defined as the maximum acceleration along the y-axis and calculated for each step. Mean PTA was determined by averaging every step taken during the 1-minute trial, then normalized to gravitational equivalents (g) to account for acceleration due to gravity (g = −9.81 m/s²).

GRF data were processed with Visual3D software (C-Motion, Inc., Germantown, MD, USA) and trimmed to include 10 acceptable foot contacts. An acceptable foot contact was defined by having one foot on each belt during double limb support. Heel-strike and toe-off events were identified from trimmed GRF data that were low-pass filtered using a bidirectional 4^th order Butterworth filter with a cut-off frequency of 12 Hz. The vertical impact peak (VIP) was defined as the maximum vertical force between heel-strike and mid-stance of the vertical GRF waveform. The vertical average loading rate (VALR) and vertical instantaneous loading rate (VILR) were calculated as the average and peak slope of the most linear portion from heel-strike to VIP.³⁷ The VIP, VALR, and VILR were averaged across all 10 steps.

2.4. Statistical Analysis

Data from the index knee were analyzed for the current study. Continuous variables were presented as the mean ± standard deviation (SD). Data were checked for normality with Shapiro-Wilkes tests. Influential outliers were removed based on visual inspection of scatterplots. Pearson’s product-moment correlations were calculated to assess the relation of PTA to measures of knee pain and impact loading. Separate multiple linear regression models were used to assess the relation of PTA to each measure of impact loading, accounting for gait speed and knee pain. Predictors were added iteratively, starting with PTA, to assess the change in coefficient of determination (R²) with the addition of each variable. Prior to regression analyses, PTA, knee pain, and gait speed were investigated for collinearity by calculating variance inflation factors (VIFs). Model assumptions were assessed with q-q plots and scatterplots of residuals and predicted values. Results were considered statistically significant at p<0.05. All statistical analyses were performed with SAS software (version 9.3, SAS Institute, Cary, NC, USA).

3.0. Results

Forty-two adults were enrolled, and their characteristics are summarized in Table 1. IMU and VAS data were available in all 42 participants. Impact loading data were available in 40 of 42 participants; data were deemed missing or invalid due to lack of acceptable foot contacts for 2 participants. Three influential outliers were removed from the impact loading data for Pearson’s analyses. All model assumptions were met for the regression analyses and no multicollinearity was found (all VIFs < 1.3).³⁸ Two influential outliers were removed for the models involving VALR and VILR, and four influential outliers were removed for the model involving VIP.

Table 1.

Demographics and clinical characteristics of participants with clinically defined knee OA (n=42)

Age, years	60.4 ± 12.6
Female, %	66.7
Height, cm	168.4 ± 9.8
Mass, kg	85.1 ± 20.0
Body mass index, kg/m2	30.0 ± 6.8
Treadmill gait speed, m/s	1.1 ± 0.2
Peak vertical tibial acceleration, g	2.4 ± 0.4
Knee pain severity, 0–100 mm Visual Analog Scale	50.3 ± 20.1
Vertical impact peak, BW	1.1 ± 0.1
Vertical average loading rate, BW/s	9.5 ± 2.4
Vertical instantaneous loading rate, BW/s	12.8 ± 3.1

Variables are described as mean ± standard deviation unless noted otherwise

The correlation results for PTA, knee pain, and impact loading are shown in Table 2. Higher PTA was significantly associated with worse VAS knee pain (r=0.39). PTA was not found to be significantly associated with VIP (r=0.28). The strongest associations were found between PTA and vertical load rates. Higher PTA was significantly associated with VALR (r=0.40) and VILR (r=0.46).

Table 2.

Correlations between peak vertical tibial acceleration, knee pain, and impact loading variables

	PTA
	Correlation coefficient	p-value
VAS, 0–100 mm	0.39	0.01
VIP, BW	0.28	0.10
VALR, BW/s	0.40	0.01
VILR, BW/s	0.46	0.004

VAS, Visual Analog Scale

VIP, Vertical Impact Peak

VALR, Vertical Average Loading Rate

VILR, Vertical Instantaneous Loading Rate

After adjusting for gait speed and pain, multiple linear regression analyses showed that PTA was a significant predictor of vertical load rates, but not VIP (Table 3). Approximately 5% of the variability of VIP was explained by PTA alone. Gait speed and pain explained an additional 28% and 4% of variability in VIP, respectively. Every 1.0 g increase in PTA accounted for a 0.06 BW increase in VIP, after adjusting for gait speed and pain. Approximately 12% of the variability of VALR was explained by PTA alone. Gait speed and pain explained an additional 15% and 6% of variability in VALR, respectively. Every 1.0 g increase in PTA accounted for a 2.68 BW/s increase in VALR, after adjusting for gait speed and pain. Approximately 27% of the variability of VILR was explained by PTA alone. Gait speed and pain explained an additional 1% of variability in VILR. Every 1.0 g increase in PTA accounted for a 3.71 BW/s increase in VILR, after adjusting for gait speed and pain.

Table 3.

Multiple linear regression analyses of peak vertical tibial acceleration, gait speed, and knee pain as predictors for impact loading variables

Dependent Variable; Full Model Results	Predictors	Unstandardized β coefficient	Standard Error	R2 Change	95% CI	p-value
VIP R2 = 0.36 F(3, 36) = 6.19, p = 0.001	PTA	0.06	0.04	0.05	−0.02, 1.13	0.13
	Gait speed	0.28	0.08	0.27	0.10, 0.45	<0.01
	Pain	<0.00	<0.00	0.04	−0.00, 0.01	0.14
VALR R2 = 0.33 F(3, 36) = 5.49, p = 0.003	PTA	2.68	1.05	0.12	0.54, 4.81	0.02
	Gait speed	5.79	2.24	0.15	0.83, 10.75	0.02
	Pain	−0.03	0.02	0.06	−0.07, 0.01	0.09
VILR R2 = 0.28 F(3, 34) = 3.88, p = 0.02	PTA	3.71	1.19	0.27	1.28, 6.13	<0.01
	Gait speed	1.29	3.03	<0.01	−4.88, 7.47	0.67
	Pain	−0.00	0.02	<0.01	−0.04, 0.04	0.86

VAS, Visual Analog Scale

VIP, Vertical Impact Peak

VALR, Vertical Average Loading Rate

VILR, Vertical Instantaneous Loading Rate

4.0. Discussion

The purpose of this cross-sectional study was to determine the relation of PTA to knee pain and impact loading during walking in adults with clinically defined knee OA. Our results suggest that an individual who walks with a higher PTA may have higher knee pain and higher vertical load rates. In addition, we found PTA to account for 28–33% of the variability in vertical load rates after adjusting for gait speed and knee pain. Thus, PTA may be an acceptable surrogate measure of impact loading that can be used in clinical environments, and in the development of rehabilitation treatments to manage symptoms of knee OA.

Previous investigations reporting PTA during walking in the knee OA population are limited. Turpin et al examined PTA of individuals with medial tibiofemoral OA before inserting shock-absorbing insoles in their usual shoes.³⁹ Christiansen et al measured PTA of individuals with end-stage knee OA during the first and last minutes of a 6-minute walk test, within 2 weeks of receiving unilateral knee arthroplasty.⁴⁰ Together these studies reported mean PTAs are approximately 1 SD (0.4 g) lower compared to our participants. Methodology differences in the accelerometer attachment and placement methods, as well as the gait analysis, may explain the higher PTA that we reported. In the prior studies, skin-mounted accelerometers were attached to the tibial tuberosity, compared to the current study which used an IMU placed above the medial malleolus. In addition, the gait analysis in the prior studies were assessed over 5 overground walking trials³⁹ and 8 steps from a 6-minute walk test,⁴⁰ compared to 60 seconds of treadmill walking in the current study. Since peak accelerations are higher with distal accelerometer placement compared to proximal,⁴¹ and during treadmill running compared to overground,⁴² the difference in PTAs reported are both reasonable and expected.

The association between PTA and VAS knee pain identified in the current study are consistent with a previous study that examined PTA during overground walking in knee OA patients, before and after pain relief.⁴³ In this study, PTA and VAS pain significantly decreased from 2.0 g and 37 mm to 1.76 g and 1.8 mm, respectively, after local analgesia (10 ml lidocaine) was injected into the knee joint. This supports the notion that higher PTA is related to greater pain. With regards to the association between PTA and impact loading, results from the current study are similar to previous reports in runners. For instance, Tenforde et al reported moderate to strong correlations between PTA and vertical load rates in injured runners (r = 0.66–0.82),⁴⁴ while others have reported slightly stronger correlations in healthy runnerPs (rre=-p0.r7o0o-0f.95).^23,24,26 Variability in the strength of the correlation can be explained by several non-standardized factors, including running speed, fatigue, footstrike pattern, accelerometer attachment methods and placement, and running surface.^45,46 As indicated in our regression models, some variance in vertical load rates remains unexplained by PTA, gait speed, and knee pain. Thus, further research is needed to identify the factors relevant to knee OA.

Our findings highlight the clinical applicability of using PTA as a surrogate measure for impact loading, particularly vertical load rates. There is growing interest in the development of treatments utilizing gait modification strategies that aim to address the aberrant impact loading characteristics and pain that manifests from knee OA. Previously reported gait modification interventions, though effective, have used methods cannot be easily implemented outside of a laboratory setting.¹¹ In these interventions, reductions in the external knee adduction moment were obtained by modifying foot progression angle, increasing medial knee thrust, and increasing trunk lean with real-time biofeedback delivered via motion capture technology and force plates. Alternatively, commercially available IMUs offer the opportunity to simultaneously deliver real-time biofeedback and assess impact loading, by proxy of PTA, in a clinical or free-living environment using significantly less time and resources.

Prior gait modification interventions in runners have demonstrated that reductions in PTA is a motor pattern that can be taught and learned.^29,30,47 Whether individuals with knee OA can feasibly reduce PTA while walking without exacerbating symptoms is unknown but holds potential as a treatment strategy. If the magnitude or rate of impact loading is reduced, then the amount of load transmitted to the knee joint may also reduce. In the absence of normative PTA data for knee OA, future research can also consider who may benefit most from this strategy (e.g., treatment responders and non-responders), and what percent reduction in PTA is both feasible and clinically meaningful. Using data from the current study as an example, individuals with mean PTA ≥ 2.4 g (i.e., 50^th percentile) may be “treatment responders” because they have the most room to reduce PTA, compared to the individuals with the lowest mean PTA.

The results should be interpreted with consideration of several limitations. As with any study, there were a number of unmeasured factors (e.g., joint malalignment, type of OA and severity) that may introduce bias to the strength of the correlations reported. These variables may be considered in future research. Our gait analysis was performed on a treadmill and whether our findings transfer to outdoor or overground walking remains unknown. Additionally, the current study examined PTA and impact loading in the index limb only because patients typically seek clinical care for the symptomatic knee. It is conceivable that the associations between PTA and impact loading may differ if taking bilateral pain into consideration. Finally, the VAS is a more global assessment of knee pain that does not measure specific locations or patterns of pain. Similarly, both PTA and the impact loading variables are more representative of lower extremity movement patterns and not knee-specific measures. However, knee-specific measures require costly and cumbersome measurement techniques in comparison to measuring PTA with commercially available wearable devices.

5.0. Conclusion

This is the first study to characterize the relationship between PTA, knee pain, and impact loading metrics during walking in knee OA. Our analyses demonstrate that PTA is significantly associated with knee pain and vertical load rates, but not the vertical impact peak. Given the potential for clinical implementation, future work examining the effects of modifying PTA on pain and impact loading is warranted.

Highlights.

• Pain and aberrant loading while walking is present in knee osteoarthritis

• Tibial acceleration was explored as a clinically accessible measure for loading

• Higher peak tibial acceleration was associated with worse knee pain

• Higher peak tibial acceleration was associated with higher vertical load rates