The Immediate Effect of Backward Walking on External Knee Adduction Moment in Healthy Individuals

Backward walking (BW) has been recommended as a rehabilitation intervention to prevent, manage, or improve diseases. However, previous studies showed that BW significantly increased the first vertical ground reaction force (GRF) during gait, which might lead to higher loading at the knee. Published reports have not examined the effects of BW on medial compartment knee loading. The objective of this study was to investigate the effects of BW on external knee adduction moment (EKAM). Twenty-seven healthy adults participated in the present study. A sixteen-camera three-dimensional VICON gait analysis system, with two force platforms, was used to collect the EKAM, KAAI, and other biomechanical data during BW and forward walking (FW). The first (P < 0.001) and second (P < 0.001) EKAM peaks and KAAI (P=0.02) were significantly decreased during BW when compared with FW. The BW significantly decreased the lever arm length at the first EKAM peak (P=0.02) when compared with FW. In conclusion, BW was found to be a useful strategy for reducing the medial compartment knee loading even though the first peak ground reaction force was significantly increased.


Introduction
Te knee is one of the most important load-bearing joints during daily activities. During level walking, the forces across the knee are not distributed symmetrically between the medial and lateral compartments; rather, over 60% of the loading goes through the medial compartment of the knee. Tis asymmetry in the distribution of the force leads to a higher medial knee loading during gait [1]. Based on the fnding previously reported, increased force is one of the potential contributing factors to the development and progression of some musculoskeletal diseases [2]. However, it is difcult to measure the loading at the knee during daily activities and it has been widely accepted that external knee adduction moment (EKAM) is a surrogate measure of the medial compartment load [3,4].
A previous study indicated that every one unit (% Nm/ Bw * Ht) increase in EKAM would result in a reduction of 0.63 mm in the knee joint space width [5]. Moreover, a published report showed that every one-unit increase in EKAM is associated with a 6.46 times increase in the risk of progression of medial compartment knee osteoarthritis (OA) [6]. Terefore, reducing the EKAM has become the objective of gait modifcation in an attempt not only to reduce the medial knee loading but also to attempt in arresting disease progression [7]. In addition to EKAM, the knee adduction angular impulse (KAAI) has also been reported as a predictor of knee loading in the medial compartment [8,9]. Moreover, some previous studies indicated that the KAAI is a more sensitive indicator than EKAM, as the frst and second EKAM peaks during gait just represent the peak knee joint loading at the early and late stance phase, while the KAAI represents the entire loading in the medial compartment of the knee joint during gait [8,9]. Terefore, it is important to include the KAAI in evaluating the loading of the knee during gait.
Backward walking (BW) has been reported as an effective rehabilitative exercise for improving the equilibrium of the human body, knee proprioception, and physical function [10][11][12][13][14]. It has been frequently recommended as a treatment for individuals who sufer from stroke, Parkinson's disease, and lower limb joint diseases [10,12,15]. Chen et al. [12] demonstrated that BW could signifcantly improve pain symptoms, physical function, and static stability in individuals with knee OA. Moreover, Gondhalekar and Deo [16] reported that BW could help to improve the clinical symptoms, knee joint range of motion (ROM), and hip abductor and extensor strength in individuals with knee OA.
Some previous studies indicated that the BW showed signifcantly higher frst vertical ground reaction force (GRF) in the early stance phase, even though the walking speed in BW was statistically lower than that in FW [17][18][19]. Te positive correlation between the frst EKAM peak and frst vertical GRF has been reported by some previous studies [20,21]; therefore, the BW might lead to higher loading (EKAM) at the medial compartment of the knee during gait. However, no previous study reported the efects of BW on medial compartment knee loading during gait, and whether changes in walk direction could afect the EKAM and KAAI has not been studied. Te aims of this study, therefore, were to investigate the efects of BW on EKAM in healthy adults. Since EKAM is the product of the GRF and the EKAM lever arm [22], this study also examined the changes in GRF and EKAM arms as secondary outcomes.

Participants.
Twenty-seven healthy adults (15 men and 12 women) with a mean age of 25.04 ± 5.44 years, mean height of 1.69 ± 0.09 m, and mean body mass of 63.91 ± 12.81 kg participated in the present study (Table 1). Participants were recruited from the postgraduate students and staf of the Shanghai University of Traditional Chinese Medicine (SUTCM). Participants who met the following criteria were included in the study: (1) healthy; (2) 18 years and older; (3) no history of injuries to the lower limbs; (4) no disease or lower limb deformities that would afect gait patterns; (5) and able to walk without any assistive devices or aids. Participants who had any musculoskeletal or neurological disorders were excluded from the study.
Before formal data collection, the purpose of this study and the procedures involved were fully explained to each participant, and written informed consent was signed before their enrollment. Based on a statistical design, the sample size for this study was calculated using the software G * Power (Version 3.1.9.6, University of Kiel, Germany), using an F-test statistical design (for EKAM) repeated measures with an efect size of 1.01 reported by a previous study, sample power of 95%, and an alpha value of 0.05. Te analysis showed that a sample size of at least 9 participants would be suitable [3].

Ethics Statement.
All experiments were approved by the China Ethics Committee of Registering Clinical Trials (ChiECRCT-20170066).

Instrumentation.
Te gait test was performed in the gait lab of Shuguang Hospital with a walkway of 8.6 meters long and 6.5 meters wide covered with a timber wooden foor. Retro-refect marker's motion data were collected with the 16-camera VICON motion capture and analysis system (VICON, Oxford, UK) at a sampling rate of 100 Hz, Te ground reaction forces were recorded with two 400 × 600 mm AMTI force plates (OR6-6, AMTI, USA) integrated and synchronized with VICON system at a sampling rate of 1000 Hz. To minimize the infuence of footwear, all participants were asked to walk barefoot in BW and FW conditions.
Refective markers (14 mm diameter) were attached to participants' specifc bony landmarks (anatomical markers) according to a previously established model [23]. Tese were located at the anterior superior iliac spine (ASIS), posterior superior iliac spine (PSIS), iliac crest, greater trochanter, medial/lateral femur epicondyle condyles, lateral/medial malleolus, 1st, 2nd, and 5th metatarsal heads and calcaneus of both limbs. Te markers placed on the anatomical landmarks were used to defne the local coordinate system and joints center of each segment. Five rigid marker cluster plates with four noncollinear markers on each of them were used to track the movement of the pelvis, thigh, and shank segment and were fxed with an elastic bandage (Fabrifoam, USA) on the anterofrontal aspect of the bilateral shank, thigh and around the pelvis. Te Calibrated Anatomical System Technique (CAST) was employed to determine the trajectory of these rigid segments and anatomical signifcance during the dynamic trials [24]. Te test-retest reliability for measuring the variables of interest (i.e., 1stEKAM, 2ndEKAM, KAAI, and walking speed) in our lab was from good to excellent (intraclass correlation coefcient (ICC) of 0.83-0.95) in twelve healthy participants.

Data
Collection. Te VIOCN system was calibrated before the participants arrived. Upon the participants' arrival, the details of the experiment were explained, and consent forms were distributed and obtained after providing enough time for participants to think, ask questions, and decide. Before the formal gait test, participants' demographic characteristics (age, height, weight) were measured. Ten, participants were asked to change into their shorts and a comfortable t-shirt. A static trial was performed for each condition before performing walking trials. Participants were instructed to walk backward with their head and upper body in a natural straight position. Both FW and BW trials were performed in one session. To refect the natural stride length, the participants were asked to walk at their selfselected speed and comfortable pace. Moreover, the participants were asked to perform ten valid trials in each of the two conditions in a randomized order [17]. Te sequence of FW and BW was randomly decided by asking participants to draw a card from a box; each card had a diferent sequence. Before formal data collection, each participant was given a few minutes for practice to get familiar with the way of walking and 20 minutes between conditions was given to allow participants to rest to minimize the fatigue efect. Considering the difculties of BW and the possible fatigue efect on the biomechanical outcomes, participants were allowed to have a 20-second short break between trials. Te kinematic and kinetic data were presented in a stance phase that was normalized to 100%. Te data normalization has been added in the data results and analysis part. GRF was normalized to body weight and the EKAM was normalized to the participant's body mass.

Data
Analysis. VISUAL 3D (Version 6.01.16, C-motion, USA) was used to calculate the kinematic and kinetic outcome measures. Te kinematics data and analog data were fltered using a Butterworth 4 th -order digital flter with a cutof frequency of 6 Hz for the kinematics and 25 Hz for the analog data [24]. An inverse dynamics algorithm was used to calculate the primary biomechanical outcome, EKAM, and KAAI, in the stance phase for the trials at both BW and FW conditions. Based on a previous study [25], the quality checks were performed in Visual3D in order to avoid extreme kinetics and kinematics values. Te EKAM and KAAI were normalized to the participant's body mass (Nm/kg and (Nm/kg)•s, respectively) and both the frst and second peaks of EKAM were presented. Te frst EKAM peak was defned as the peak EKAM in the frst half of the stance and the second EKAM peak was defned as the peak EKAM in the second half of the stance. Te KAAI was defned as the positive area under the EKAM-time graph. Te KAAI incorporates both the mean magnitude of the EKAM and the time for which it is imposed on the knee. Te KAAI was calculated based on the EKAM and was presented with a unit of (Nm/kg) ·s, which was calculated by integrating the EKAM signal during a stance phase. Te knee joint moment arms were defned as the perpendicular distance between GRF and the knee joint center in the laboratory frontal plane. Calculated at the time of the frst and second EKAM peaks. Medial-lateral GRF was defned as a medial-lateral component of the GRF. Te positive value is medial force. Te early stance medial-lateral GRF was defned as the peak medial-lateral GRF in the frst half of the stance and the second peak of medial-lateral GRF was defned as the peak medial-lateral GRF in the second half of the stance. Vertical GRF was defned as the vertical component of the GRF. Te positive value is vertical force. Te frst peak of vertical GRF was defned as the peak vertical GRF in the frst half of the stance and the second peak of vertical GRF was defned as the peak vertical GRF in the second half of the stance. Te center of pressure (COP) was the distance of the center of pressure from the long axis of the foot (calcaneus to the 2nd metatarsal), where positive values indicate lateral. Calculated at the time of peak EKAM [26]. Based on a previous study, the quality checks were performed in Visual3D in order to avoid extreme data [25].
2.6. Statistical Analysis. Te data from both legs were analyzed to satisfy the independence assumption of statistical analysis. Shapiro-Wilk tests were used to assess the normality of the selected parameters. Te One-way repeated measure analysis of variance (ANOVA) was used to examine the diference in EKAM, EKAM arm, KAAI, GRF, and COP between FW and BW conditions. All statistical analyses were performed in SPSS (Version 16.0, IBM Corporation, USA) with a global alpha level of 0.05. For those variables that were signifcantly diferent between FW and BW, the relationships between their mean diferences and mean diference in the frst EKAM peak were evaluated using the Pearson r correlation coefcient [26]. Correlation coefcients were assumed very high when R > 0.8, high when R � 0.6-0.8, medium when R � 0.4-0.6, weak when R � 0.2-0.4, and very weak when R < 0.2 [19].

Results
Te Shapiro-Wilk tests showed that both kinematic and kinetic parameters were normally distributed (P > 0.05). Te biomechanical data in each condition are shown in Table 2.
A signifcant reduction in walking speed (1.30 ± 0.10 m/s, vs. 0.97 ± 0.13 m/s, P < 0.001) was observed in BW when compared with FW due to the invisibility of the direction of progress, however, the frst peak vertical GRF in BW was signifcantly higher when compared with FW (P < 0.001) (Table 2, Figure 1). BW reduced the frst EKAM peak, second EKAM peak, KAAI, frst peak EKAM arm, and second peak vertical GRF signifcantly by 26.3% (P < 0.001), 16.0% (P � 0.008), 16.7% (P � 0.02), 40.0% (P � 0.02), and 15.0% (P < 0.001), respectively ( Table 2, Figures 1 and 2). Moreover, the medial-lateral GRF was found to act in the opposite direction in two conditions. For FW the medial-lateral GRF was positive, the GRF acted medially, while it acted laterally (negative) in BW (P < 0.001, Table 2, Figure 3). However, there was no signifcant diference in the second peak EKAM arm between FW and BW (P � 0.25) ( Table 2). When compared with FW, BW signifcantly shifted the COP medially at the time of the frst EKAM peak (P < 0.001), and no signifcant diference in COP at the time of the second EKAM peak between FW and BW was found (P > 0.05) ( Table 2). Te relationships between mean change in variables that were signifcantly diferent between FW and BW in early stance and the frst EKAM peak are reported in Table 3. Mean change in the frst peak EKAM arm (r � 0.50), KAAI (r � 0.58), and COP at the time of the frst EKAM peak (r � −0.43) was signifcantly covariate with the frst EKAM Journal of Healthcare Engineering peak (Tables 2 and 3). Tus, the frst EKAM peak was strongly infuenced by the frst peak EKAM arm, KAAI, and COP at the time of the frst EKAM peak. Tere was no statistically signifcant relationship between the mean change in walking speed (r � 0.04) and the mean change in the frst EKAM peak (Tables 2 and 3). Tus, the frst EKAM peak was poorly infuenced by walking speed in this study.

Discussion
Some previous studies in BW focusing on lower limb joint diseases were performed on subjects with the specifed problems [10][11][12][13][14]. However, the frst vertical GRF in BW has been proven to be signifcantly higher in the early stance phase when compared with FW, even though the walking speed in BW was statistically lower than that in FW [17][18][19], which might lead to higher loading at the knee during gait.
To date, no previous study reported the efects of BW on medial compartment knee loading during gait, and whether changes in the frst peak GRF could afect the EKAM and KAAI has not been studied. Although there are indeed certain diferences in biomechanical outcomes between patients and healthy individuals. Tis study was performed on healthy subjects because the outcomes would improve our understanding of the biomechanical impact of BW on knee loading, which has not been well-defned in previous studies. Terefore, no subjects with specifed lower limb problems were employed for the study.
Our fndings showed that BW signifcantly reduced the frst and second EKAM peaks and KAAI. Such observations may be explained by the reduction of the EKAM arm caused by the opposite direction of medial-lateral GRF in early stance and the reduction of second peak vertical GRF during BW when compared with FW. 0.06 ± 0.01 -0.08 ± 0.02 −0.14 ± 0.03 <0.00 * * Late stance medial-lateral GRF (body weight) 0.05 ± 0.02 -0.05 ± 0.01 −0.10 ± 0.03 <0.00 * * Te frst peak of vertical GRF (body weight) 1.15 ± 0.09 1.32 ± 0.14 0.16 ± 0.13 <0.00 * * Te second peak of vertical GRF (body weight) 1.13 ± 0.08 0.96 ± 0.07 −0.19 ± 0.08 <0.00 * * COP at the time of the frst EKAM peak (mm) 7.73 ± 6.72 3.99 ± 5.72 −3.74 ± 9.00 0.004 * COP at the time of the second EKAM peak (mm) 7.66 ± 7.67 6.51 ± 6.19 −1.15 ± 9.47 0.37 * P < 0.05; * * P < 0.001. Values were the mean ± SD. FW � forward walking, BW � backward walking, MD � mean diference, EKAM � external knee adduction moment, KAAI � knee adduction moment impulse, GRF � ground reaction force, and COP � center of pressure. A greater COP value refers to a more laterally located COP.   Journal of Healthcare Engineering Compared with FW, the walking speed in BW was reduced by 25.3%. Tis fnding was in accordance with a previous study that reported a 15.3% reduction in walking speed in BW [17]. Tis could be explained by a more cautious gait strategy due to the invisibility of the direction of the progress [27]. It was generally reported that decreased walking speed was associated with lower GRF during levelground walking [28,29]. However, the BW showed a larger frst peak GRF during the early stance phase, which was similar to previous studies [17,19]. Te mean diference in walking speed and frst peak vertical GRF were not signifcantly correlated with the mean diference in the frst EKAM peak between FW and BW. Tis result indicated that the decrease in the frst EKAM peak might not be dependent on the decreased frst peak of vertical GRF caused by the reduction of walking speed. A similar fnding has been reported by one previous study [19], which also demonstrated that the GRF peaks were poorly infuenced by walking during BW. Moreover, we found the mean diference in the frst peak EKAM lever arm was signifcantly correlated with the mean change in the frst EKAM peak, which indicated that the reduction of EKAM was caused by the change in the EKAM lever arm. Te second peak vertical GRF in BW was also signifcantly smaller than that of FW. Tese fndings were similar to previous studies [17][18][19]. Since BW showed signifcantly higher frst peak vertical GRF when compared with FW, the reduction of the frst EKAM peak could be due to the shortening of the moment arm in BW, as the EKAM is mainly determined by the magnitude of GRF at the frontal plane (i.e., GF frontal ) and the perpendicular distance (EKAM arm) between the knee center and the GRF vector [30].
Previous studies indicated that the EKAM arm reduction could be explained by the lateral shift of the COP due to the shortening of the perpendicular distance between the GRF vector and the knee joint center [31,32]. However, BW showed a signifcantly smaller frst EKAM peak even though the COP at the time of the frst EKAM peak was shifted medially when compared with FW. Te medial shift of COP at the frst EKAM peak can be explained by the toe-contact to heel-contact during the early stance in BW [17], as a previous study indicated that the COP was shifted more medially during heel-of to toe-of in comparison with heelstrike to foot-fat in FW [13]. Since there was a signifcant medial shift of the COP in BW during the early stance, we suggested that other factors rather than the shift of the COP might have contributed to the decrease in the frst peak EKAM arm in BW. We found that the magnitude of mediallateral GRF in BW was quite diferent from that of FW. In FW, the magnitude of the early stance medial-lateral GRF was positive (medial direction) whereas BW showed negative early stance medial-lateral GRF (lateral direction) ( Figure 3). Tis fnding indicated that the direction of the medial-lateral GRF in BW was opposite to that of FW. Te calculation of the frontal plane component of GF (i.e., GF frontal ) was dependent on the GF mediolateral and G⟶RF vertical [30]. Due to the changes of GF mediolateral , the perpendicular distance between the knee joint center and the frontal plane component of GF frontal was decreased (Figures 4  and 5), consequently, the EKAM arm was decreased so that the signifcantly smaller frst EKAM peak was achieved in BW in comparison with that of FW, which was the real reason for EKAM reduction in BW. No signifcant diference in the second peak EKAM arm and the COP at the time of the second EKAM peak was found. However, the second EKAM peak was still signifcantly reduced by BW when compared with FW. Tis might be explained by BW signifcantly reduced vertical GRF by 15.0% in the second peak.
Additionally, the KAAI in BW condition was signifcantly reduced, which indicated that BW could not only reduce the EKAM peak, but also the whole loading of the knee during the stance phase. Previous researchers believed that the KAAI was more important than the peak value of EKAM, as it represents the resultant loading efect over the stance time in the medial compartment of the knee joint [8]. However, more studies concluded that the peak value of EKAM especially the frst EKAM peak was strongly associated with the knee OA progression and was the main reason causing pain and damage to the knee joint [5,33,34]. Tus, the reduction in the EKAM peaks and KAAI in BW could be regarded as the  Journal of Healthcare Engineering 5 positive outcome of the walking exercise for relieving the loading at the knee during gait. EKAM peaks and KAAI were shown to be positively correlated with the development and progression of some musculoskeletal disorders such as knee OA [6]. Te current study showed that BW could efectively reduce the EKAM peaks and KAAI. Terefore, we hypothesized that BW may be a useful strategy for individuals with medial compartment knee OA. With our preliminary fndings among healthy young adults, such hypotheses should be tested in individuals with knee OA before conducting another randomized controlled trial.
Tere were several limitations in the current study. Firstly, only healthy young adults were recruited in the current study. Our fndings should be further examined on the individual with medial compartment knee OA. Secondly, no surface electromyography (sEMG) data of the lower limb was included to identify the muscle activity in BW as previous studies have shown that muscle co-contraction infuences knee joint compressive loading [35,36]. However, healthy individuals are unlikely to have aberrant muscle cocontraction and thus this would be recommended in individuals with medial knee OA. Finally, lateral trunk lean was not measured and analyzed in the current study. Although in the protocol each participant was advised to keep their head and upper body in a natural straight position and both upper limbs waving in a natural way during the BW test. Te efect of trunk posture on knee loading during gait in individuals with knee OA is critical [37]. Terefore, the outcome of lateral trunk leans during BW should be analyzed in the future.

Conclusion
Te results of our study confrmed the signifcant biomechanical function of BW in reducing the knee joint loading, with the frst and second EKAM peaks and KAAI decreasing signifcantly due to the change of the medial-lateral GRF and the reduction in the EKAM arm. We suggested that BW may potentially be a viable strategy in reducing medial knee loading for individuals with medial compartment knee OA.

Data Availability
Te data to support the fndings of this study are available from the corresponding author upon reasonable request.

Disclosure
Jian Pan and Jiehang Lu are the co-frst authors. Hongsheng Zhan and Meng Kang are the co-corresponding authors.