Gait differences after ACL reconstruction versus healthy controls: a three-level meta-analysis of kinematics, kinetics, and spatiotemporal parameters Authors: Shiwei Chen, Wenling Gou, Renzhi Gao Published: 2025 ABSTRACT AbstractObjectivesThis meta-analysis was designed to evaluate whether individuals following ACL reconstruction exhibit normalized gait biomechanics when compared to healthy controls.MethodsThis review was conducted across five databases: PubMed, Embase, The Cochrane Library, Web of Science and EBSCOhost-SportDiscus, yielding a total of 5,251 records, with 21 qualified studies (52 independent samples from 861 participants; based on available data, the mean age was 26.5 ± 4.3 years and 51.3% were female) eligible for quantitative synthesis following rigorous screening. A three-level meta-analysis model was implemented to address effect size dependencies. Six gait parameters were analyzed, including kinematic variables (peak knee flexion and extension angles during stance), kinetic variables (peak external knee flexion and extension moments during stance) and spatiotemporal variables (stride length and walking speed).ResultsAmong the six gait parameters that were analyzed, only the peak knee extension angle during stance (k = 9, d = 0.83, 95% CI: 0.26–1.40, I2 = 77.53%, p = 0.01) showed a significant difference between ACL reconstruction patients and healthy controls. However, peak knee flexion angle, peak external knee flexion and extension moments during stance, stride length and walking speed were not significantly different between groups (p > 0.05). Meta-regression identified a significant time-dependent improvement only for knee extension angle (F = 5.656, p = 0.049), while postoperative duration had no significant effect on other outcomes.ConclusionThis meta-analysis shows that the majority of gait parameters during stance, including peak knee flexion angle, external knee flexion and extension moments, stride length and walking speed, do not differ significantly between individuals with ACL reconstruction and healthy controls. Only peak knee extension angle remained impaired, though it showed signs of improvement over time. Persistent knee extension deficits emphasize the need for targeted quadriceps and knee control training as a key focus for long-term ACLR rehabilitation.Keywords: Postoperative recovery, gait adaptation, three-level modelling, functional recovery, long-term rehabilitation Introduction Anterior cruciate ligament (ACL) injuries rank among the most prevalent sports-related injuries, particularly in high-intensity sports activities, resulting in significant impairment to adolescents’ motor function and interfering with routine activities [1–3]. Statistics show that around 250,000 ACL injuries occur each year in the United States, with approximately 30% of patients suffering secondary injuries within 24 months of the initial injury [4]. Furthermore, 45–50% of ACL-injured individuals develop knee osteoarthritis within 10–15 years [5,6]. Anterior cruciate ligament reconstruction (ACLR) surgery remains the primary treatment for ACL injuries, a procedure that is ultimately accepted by approximately 65% of patients [7,8]. Although ACLR contributes to restoring knee joint stability [9], studies have revealed that compared to healthy controls, the ACL-injured patients undergoing reconstruction still exhibit abnormal lower limb biomechanical patterns in gait performance, including aberrant walking kinematics [10,11]. Prior data also demonstrate reduced knee extension moments and altered knee angles in postoperative ACLR populations [12–14]. However, these data on gait restoration after ACLR are equivocal. While some studies suggest that postoperative patients can regain near-normal gait patterns within a short period of time [15], other data indicate that gait restoration follows a prolonged progressive trend towards normative levels [16,17]. These discrepancies may reflect heterogeneity in severity of the ACL injury, postoperative follow-up duration, and the content or effectiveness of rehabilitation protocols and surgical techniques across cohorts. Consequently, despite extensive investigations into altered joint loading patterns in gait after ACLR, a systematic understanding of long-term postoperative gait recovery remains elusive [18]. Therefore, it is critical to integrate existing evidence through systematic analysis to explore long-term changes in gait recovery. Meta-analysis, as a powerful statistical tool, facilitates the integration of findings from numerous studies, improves statistical efficacy and provides more precise effect estimates to advance insights into gait changes after ACLR [19]. However, the existing meta-analysis studies on gait after ACLR have not been updated for a period of 7 years, with persistent limitations in terms of methodology and conclusions as outlined below. Most of the contemporary meta-analyses often relies on the assumption of an independent study effect, where only a single effect size is extracted from each study [12,13,20,21]. However, the gait assessment often involves multiple time points (e.g. 3, 6 and 12 months after ACLR) and multiple outcome measures (e.g. knee flexion angles, knee extension moments) [18,22,23]. These methodological constraints may lead to the loss of information, thereby fundamentally limiting the capacity to fully reveal the long-term process of gait recovery. Given the limitations of existing meta-analyses in the research of gait recovery after ACLR, this study employs a three-level meta-analysis approach to overcome constraints of the single effect size assumption inherent in conventional meta-analysis and to fully incorporate data from multiple time points and outcome measures involved in gait assessment. It synthesizes multidimensional gait parameters, including kinematic, kinetic and spatiotemporal measures, to more accurately reveal the gait biomechanical differences between individuals with ACLR and healthy controls. The use of a three-level model allowed for the inclusion of multiple time points and outcome measures across studies, thereby enabling a more precise evaluation of whether gait biomechanics normalize after ACL reconstruction based on all available between-group comparisons. This comprehensive approach strengthens the credibility of the findings and provides clearer guidance for identifying persistent impairments and refining long-term rehabilitation strategies. Based on the existing literature, we hypothesized that although most kinematic, kinetic and spatiotemporal gait parameters would approximate those observed in healthy controls following ACLR, persistent deficits in knee extension mechanics would remain. Methods This review followed the PRISMA guidelines for systematic evaluation and meta-analysis [24]. In addition, this review has been registered in the (PROSPERO) database on April 2, 2025, with the registration number CRD420251026996. Literature search and selection To identify original studies comparing the gait of patients with ACLR and healthy individuals, a systematic search of the following databases was performed: PubMed, Embase, The Cochrane Library, Web of Science and Ebscohost-Sportdiscus. Specifically, the following terms and the combinations thereof were used in the search for titles, abstracts and keywords: ‘anterior cruciate ligament reconstruction’ (or ‘ACLR’), ‘anterior cruciate ligament’ (or ‘ACL’) and ‘gait’ (or ‘walking’). These broad terms ensured that studies assessing gait after ACLR were captured, regardless of which specific biomechanical outcomes were analyzed. In contrast, adding outcome-specific terms (e.g. ‘knee moment’, ‘knee angle’, ‘walking velocity’, ‘ground reaction force’) would have restricted the search to studies that explicitly listed these variables in the title or abstract, even though many relevant analyses report such outcomes only within the full text. The search table and the corresponding results for each database were provided in Appendix 1. Inclusion criteria To be eligible for inclusion, original studies are required to meet the following criteria: Subjects. Participants must be individuals with a primary ACL rupture who have undergone surgical reconstruction. No restrictions are imposed on comorbid injuries, rehabilitation strategies, or surgical approaches in this study. Subjects. Participants must be individuals with a primary ACL rupture who have undergone surgical reconstruction. No restrictions are imposed on comorbid injuries, rehabilitation strategies, or surgical approaches in this study. Comparison groups. Studies must include a healthy control group. In this review, ‘healthy’ refers to individuals without a history of ACL injury, other major knee injuries, or current knee pain or dysfunction. No minimum follow-up duration was required, as studies were eligible as long as participants had completed ACL reconstruction at the time of gait assessment. Comparison groups. Studies must include a healthy control group. In this review, ‘healthy’ refers to individuals without a history of ACL injury, other major knee injuries, or current knee pain or dysfunction. No minimum follow-up duration was required, as studies were eligible as long as participants had completed ACL reconstruction at the time of gait assessment. Studies comparing only the ACLR limb with the contralateral limb were excluded, as intra-individual comparisons may introduce bias due to inherent limb asymmetries, whereas healthy controls provide a more stable reference for normal gait patterns [25]. No restrictions were placed on whether studies included single or multiple postoperative measurement points, as long as at least one gait assessment after ACL reconstruction was reported. Studies comparing only the ACLR limb with the contralateral limb were excluded, as intra-individual comparisons may introduce bias due to inherent limb asymmetries, whereas healthy controls provide a more stable reference for normal gait patterns [25]. No restrictions were placed on whether studies included single or multiple postoperative measurement points, as long as at least one gait assessment after ACL reconstruction was reported. Outcome measures. To be eligible, studies were required to report at least one quantitative gait-related outcome during walking on level ground, including kinematic parameters such as peak knee flexion or extension angle during stance (degrees), kinetic parameters such as peak external knee flexion or extension moment during stance (Nm/kg), or spatiotemporal parameters such as stride length (m) or walking speed (m/s). Outcome measures. To be eligible, studies were required to report at least one quantitative gait-related outcome during walking on level ground, including kinematic parameters such as peak knee flexion or extension angle during stance (degrees), kinetic parameters such as peak external knee flexion or extension moment during stance (Nm/kg), or spatiotemporal parameters such as stride length (m) or walking speed (m/s). Exclusion criteria Studies that meet any of the following conditions shall be excluded: The absence of a healthy control group, or comparison only between the ACLR limb and the contralateral limb. The absence of a healthy control group, or comparison only between the ACLR limb and the contralateral limb. The baseline characteristics of the experimental group differ significantly from those of the healthy control group. The baseline characteristics of the experimental group differ significantly from those of the healthy control group. Only mean differences between groups are reported, and no raw data or data that could be converted into effect sizes are provided. Only mean differences between groups are reported, and no raw data or data that could be converted into effect sizes are provided. The ACLR limb is not clearly specified (e.g. only ‘left leg’ or ‘right leg’ is reported, but the injured side is not specified). The ACLR limb is not clearly specified (e.g. only ‘left leg’ or ‘right leg’ is reported, but the injured side is not specified). Non-English publications. Non-English publications. Non-peer-reviewed publications, including conference abstracts, reviews, case reports, or dissertations. Non-peer-reviewed publications, including conference abstracts, reviews, case reports, or dissertations. Data extraction and coding process To identify potential moderators that may influence study outcomes, a systematic data extraction and coding process was undertaken to extract relevant features from the original studies. The specific extraction and coding process was as follows: Step 1: Data extractionTwo reviewers (S.C. & W.G.) extracted the following key information based on a customized data extraction form: study characteristics (e.g. study design, data sources); participant demographics (including country, gender, age, BMI, postoperative time); and outcome measures (mean values and standard deviations for the experimental group and the control group). Step 1: Data extraction Two reviewers (S.C. & W.G.) extracted the following key information based on a customized data extraction form: study characteristics (e.g. study design, data sources); participant demographics (including country, gender, age, BMI, postoperative time); and outcome measures (mean values and standard deviations for the experimental group and the control group). Step 2: Development of coding schemeTo achieve the study objectives and resolve the research questions, a detailed coding scheme for the gait data was developed. The scheme specifically included the following categories of outcome measures [26,27]: kinematic parameters, including peak knee flexion and extension angles during the stance phase (degrees); kinetic parameters, including peak external knee flexion and extension moments during the stance phase (Nm/kg); and spatiotemporal parameters, including stride length (m) and walking speed (m/s). Given that different gait phases (e.g. stance phase vs. swing phase) and value types (e.g. peak vs. mean) may represent distinct biomechanical constructs and affect the comparability of results across studies, only peak values during the stance phase were extracted for analysis to ensure consistency in statistical comparability and interpretation across studies. Step 2: Development of coding scheme To achieve the study objectives and resolve the research questions, a detailed coding scheme for the gait data was developed. The scheme specifically included the following categories of outcome measures [26,27]: kinematic parameters, including peak knee flexion and extension angles during the stance phase (degrees); kinetic parameters, including peak external knee flexion and extension moments during the stance phase (Nm/kg); and spatiotemporal parameters, including stride length (m) and walking speed (m/s). Given that different gait phases (e.g. stance phase vs. swing phase) and value types (e.g. peak vs. mean) may represent distinct biomechanical constructs and affect the comparability of results across studies, only peak values during the stance phase were extracted for analysis to ensure consistency in statistical comparability and interpretation across studies. Step 3: Cross-checking and validationFollowing data extraction and coding, an independent cross-check was conducted by two reviewers (S.C. & W.G.). Any discrepancies identified during this process were resolved through team discussions, followed by a re-extraction of the relevant study data. Finally, the corresponding author (R.G.) performed a thorough review of all coded data to ensure accuracy and consistency. Step 3: Cross-checking and validation Following data extraction and coding, an independent cross-check was conducted by two reviewers (S.C. & W.G.). Any discrepancies identified during this process were resolved through team discussions, followed by a re-extraction of the relevant study data. Finally, the corresponding author (R.G.) performed a thorough review of all coded data to ensure accuracy and consistency. Risk of bias and quality of methods assessment All studies included in this review were observational studies (non-randomized controlled trials) of gait characteristics between patients with ACLR and healthy controls. To assess methodological quality, two reviewers (S.C. & W.G.) employed the Cochrane Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) to assess potential biases across seven critical domains: confounders, participant selection procedures, intervention classification, deviations from expected interventions, missing data, outcome measures and selective reporting of outcomes [28]. In cases of discrepancies, consensus was reached through consultation with a third investigator (R.G.). Statistical analysis Data synthesis and effect measures This study compared gait parameters between patients who underwent ACL reconstruction and healthy controls, and quantified these differences by calculating the pooled standard deviation, standardized mean difference (SMD; Cohen’s d), and their variances to estimate effect sizes and associated uncertainty. Equation (1) shows the calculation of the pooled standard deviation (SD) [29]: SDpooled is a weighted combination of the standard deviations of the experimental group (ACLR group) and the control (healthy) group. npost_exp and npost_ctrl are the sample sizes of the post-test data of the two groups, SDpost_exp and SDpost_ctrl are the standard deviations of the post-test gait parameters of the two groups. The weighted pooling of standard deviations is a process that takes into account the respective sample sizes of the two groups to enhance the accuracy of effect size estimation. Equation (2) presents the formula for computing the standardized mean difference [30]: dexp_ctrl represents the standardized effect size between the two groups, Mpost_exp is the mean value of the gait parameters in the ACLR group, and Mpost_ctrl is the mean value of the gait parameters in the healthy control group. The pooled SD calculated in the previous step is used for standardization, and the SMD, as a dimensionless metric, is capable of comparisons between different studies. Equation (3) shows the variance of Cohen’s d (Var_d) [31]: Vard_exp_ctrl represents the variance of Cohen’s d, used for the calculation of confidence intervals for the SMD to assess the robustness of the effect size estimate. According to the Cochrane Handbook, the effect sizes of Cohen’s d are classified as follows [32]: weak (<0.2), small (0.2–0.5), medium (>0.5–0.8) and large (>0.8). Three-level meta-analysis and heterogeneity Individual outcome measures in gait-related studies often report multiple effect sizes (ESs) [33,34] derived from diverse biomechanical parameters, including kinematics and kinetics (e.g. knee flexion, extension angles and moments), and spatiotemporal parameters (e.g. stride length, walking speed and stance phase of the gait cycle) [22,35], often assessed at multiple time points [18,36]. Due to the correlation of different effect sizes (ESs) within the same study, conventional meta-analysis approaches assume independence among ESs across studies, while violations of this assumption may lead to exaggerated results [37]. Conversely, adopting an overly conservative approach – such as including only a single ES per study or averaging correlated ESs – may obscure true study trends [38]. Therefore, this study employed a three-level meta-analysis model with statistical analyses performed by the metafor R package (version 4.3.2, R Core Team, 2022), using REML (Restricted Maximum Likelihood) as the estimator to address dependencies among the effect sizes. In addition, multiple data points are incorporated from the same outcome measures within individual studies where available, to increase the overall sample sizes and provide a more objective insight into the underlying trends [32]. Meta-regression analysis and sensitivity analysis A meta-regression using postoperative duration as a moderator was performed to evaluate the potential influence of follow-up duration on the observed effect sizes. Although the primary objective did not involve the modelling of temporal recovery, this supplementary analysis offered insight into the extent to which time contributed to between-study heterogeneity and supported the robustness of the overall findings. For sensitivity analyses, to examine the robustness of the results, these outliers were excluded and the three-level meta-analytic model was re-estimated. Additionally, the influence of each study on the overall findings was assessed by iteratively omitting individual effects at the Tier 2 level (within-study variance) and individual studies at the Tier 3 level (between-study variance) of the three-level model. Publication bias In this study, the risk of publication bias was evaluated using funnel plots and Egger’s regression test (with p > 0.05 indicating no significant publication bias) [39]. Given that individual studies may contribute multiple effect sizes in a three-level meta-analytic model, publication bias was assessed at two hierarchical levels for greater precision. At Level 2 (within-study), the symmetry of effect sizes within each study was examined, while at Level 3 (between-study), the symmetry of the overall distribution across studies was evaluated based on the mean effect size per study [40]. Certainty of the evidence This study adopted a pragmatic approach to the assessment of the certainty of the evidence without the formal implementation of the GRADE framework [41]. Given that the included investigations exclusively utilized observational study designs to compare ACLR patients and healthy controls with respect to gait-related parameters, the application of GRADE - a methodology primarily optimized for randomized controlled trials (RCTs) - would result in a lower level of evidence within the research paradigm. Therefore, the GRADE assessment was deliberately excluded to prevent this study from generating a more explanatory grading of the evidence. Results Search results and study characteristics The search incorporated documents available as of April 8, 2025, yielding a total of 5,251 records. Following a rigorous screening of titles, abstracts and keywords, 85 primary studies were identified as potential candidates for inclusion, of which 21 were finally included (three identified through meta-analysis) [5,11,16–18,22,23,33–36,42–51]. Figure 1 illustrates the search process. PRISMA flow diagram of study selection. The 21 studies incorporated in this study included 52 data points from 861 participants (mean age: 27.02 ± 4.48 years for ACLR patients versus 25.97 ± 4.06 years for healthy controls), representing 11 countries with 50.2% of participants being female. All of the included studies conducted comparative analyses of gait parameters between the ACLR patients and healthy controls, including kinematic parameters: peak knee flexion and extension angle during stance; kinetic parameters: peak external knee flexion and extension moment during stance; and spatiotemporal parameters: stride length and walking speed. Participant and study characteristics are presented in Table 1. Table 1.Characteristics of included literature.AuthorCountryN (total sample)Age (years)Sex (M/F)MeasurePost-operative time (month)Body mass (kg) / BMI (kg/m)Georgoulis, 2003Greece31C:24.7 ± 3.7E:25 ± 4C: 8M and 2FE: 19M and 2FPeak knee flexion angle during stance (degree)7C:62.1 ± 12.4kgE:69.1 ± 7.9kgPadhye, 2023USA60C:22.7 ± 2.9E:22.3 ± 4.4C: 11M and 19FE: 11 M and 19FPeak external knee flexion moment during stance (Nm/kg)Walking Speed (m/s)Stride length (m)48N/ANoehren, 2013USA40C:26 ± 5.1E:25 ± 6.2C: 20FE: 20FPeak knee flexion angle during stance (degree)62C:61 ± 5.1E:64 ± 8.2Büttner, 2024USA116C:21.6 ± 4.3E: 21.8 ± 4.7N/AWalking Speed (m/s)4,6,11C: BMI:24.5 ± 3.2E: BMI:24.2 ± 3.2Armitano-Lago, 2020USA32C:28.90 ± 6.24E:29.25 ± 6.9C: 8M and 8FE: 8M and 8FWalking Speed (m/s)107C:74.61 ± 17E:81.46 ± 14Skvortsov, 2020Russia40C:34.2E:29.7C: 14M and 6FE: 10M and 10FPeak knee flexion and extension angle during stance (degree)3,12N/ALeporace, 2016Brazil22C:27.3 ± 2.7E:33.1 ± 11.1N/AWalking Speed (m/s)Stride length (m)11C:82.1 ± 9.5E:82.1 ± 7.4Wilson, 2019USA60C:22 ± 4E:22 ± 3C: 15M and 15FE: 15M and15FWalking speed (m/s)6,12C: BMI:23.6 ± 2.1E: BMI:23.8 ± 2.2Goetschius, 2018USA38C:22.4 ± 3.2E:26.7 ± 4.4C:13M and 7FE: 12M and 6FPeak external knee extension moment during stance (Nm/kg)Peak external knee flexion moment during stance (Nm/kg)Peak knee flexion angle during stance (degree)17,103C:68.9 ± 13.1E:69.5 ± 12.7Davis, 2018USA40N/AC: 9M and 11FE: 11M and 9FWalking Speed (m/s)84C: BMI:22.4 ± 2.4E: BMI:25 ± 3.5Hadizadeha, 2016Malaysia37C:21.5 ± 1.0E:23.6 ± 5.4C: 9M and 6FE:13M and 9FPeak external knee extension moment during stance (Nm/kg)3C:61.7 ± 11.3E:66 ± 13.8Patterson, 2013Ireland34C:20.8 ± 1.17E:23.7 ± 3.12C: 17FE: 17FWalking Speed (m/s)42C:64.7 ± 7.06E:64.9 ± 9.02Milandri, 2017South Africa30C:28.6 ± 6.8E:37.4 ± 10.7C: 15ME: 15MPeak external knee flexion moment during stance (Nm/kg)60C:85.2 ± 13.8E:90.3 ± 13.5Lewek, 2002USA28C:32.2 ± 6.6E: 25 ± 7.8C: 8M and 2FE: 4M and 6FPeak knee flexion angle during stance (degree)3.3,4.8N/AHall, 2012Australia32C:26 ± 4E: 26 ± 6C: 7M and 10FE: 7M and 8FPeak external knee extension moment during stance (Nm/kg)72C:68 ± 2E:75 ± 15Karimi, 2013Iran30C:32 ± 3E: 33 ± 2.6N/APeak external knee extension moment during stance (Nm/kg)Walking Speed (m/s)Stride length (m)6N/AFerber, 2002USA20C:24.4 ± 3.1E:27.7 ± 9.1C: 5M and 5FE: 5M and 5FPeak knee flexion angle during stance (degree)3C:67.2 ± 10.7E:79.1 ± 13.8Kono, 2023Japan62C:24.07 ± 4.73E:25.91 ± 11.61C: 18M and 12FE: 12 M and 10FPeak external knee flexion moment during stance (Nm/kg)Peak knee extension and flexion angle during stance (degree)12C:63.5 ± 12.4E:64.1 ± 12.6Knoll, 2004Hungary76C:31.70 ± 4.1E:34.78 ± 6.91C: 31M and 20FE: 18M and 7FPeak knee extension angle during stanceStride length (m)4,8,12C:72.1 ± 25.2E:84.7 ± 15.6Lin, 2019USA36C:25 ± 11E:25 ± 11C: 8M and 10FE: 8M and 10FPeak external knee extension moment during stance (Nm/kg)3.7C:69.7 ± 9.1E:69.8 ± 10.9Butler, 2008USA34C:23.4 ± 5.7E:23.6 ± 5.8C: 4M and 13FE: 4M and 13FWalking Speed (m/s)64C: BMI:24.0 ± 3.2E: BMI:23.2 ± 2.4Abbreviations: E = Experimental group; C = Control group; BMI = Body Mass Index. Characteristics of included literature. Abbreviations: E = Experimental group; C = Control group; BMI = Body Mass Index. Risk of bias and publication bias The risk of bias for each included study is presented in Figure 2. Most studies exhibited low risk in domains pertaining to intervention classification, data completeness and selective reporting. However, higher levels of risk were observed for confounding bias and participant selection. Specifically, one study [49] was rated as high risk for inadequate control of confounding factors, while two studies [34,47] were rated as high risk for participant selection. Risk of bias assessment across included studies. Most studies showed low to moderate risk across domains. Regarding publication bias, Figures 3 and 4 present the outcomes of the funnel plot and Egger’s regression test. In accordance with the conventional two-level model, significant asymmetry was observed for walking speed (p = 0.0003). In contrast, the stride length (p = 0.8814), peak knee flexion angle (p = 0.4339), peak knee extension angle (p = 0.5431) and peak external knee extension moment (p = 0.9914) showed a more symmetrical distribution, with Egger’s test yielding a non-significant result, indicating no substantial evidence of publication bias. Funnel plots and Egger’s test results for assessing publication bias (two-level model). Funnel plots and Egger’s test results for assessing publication bias (three-level model). In accordance with the three-level meta-analytic model, the funnel plot for walking speed (p = 0.116), stride length (p = 0.740), peak knee flexion angle (p = 0.366), peak knee extension angle (p = 0.903) and peak external knee extension moment (p = 0.194) displayed a high degree of symmetry, with Egger’s test confirming the absence of publication bias. Funnel plot analysis was not conducted for peak external knee flexion moment due to the limited number of included studies. Main effect results Differences in kinematic parameters between patients with ACLR and healthy controls No significant difference in peak knee flexion angle during stance was observed between individuals with ACL reconstruction and healthy controls (k = 11, d = −0.09, 95% CI: −0.62 to 0.45, I2 = 71.06%, p = 0.73). Variance decomposition showed that within-study heterogeneity was negligible (Tau2Level 2 = 0.06), while substantial between-study heterogeneity remained (Tau2 Level 3 = 0.26). In contrast, a significant difference was found in peak knee extension angle during stance, with ACLR patients showing significantly greater values than healthy controls (k = 9, d = 0.83, 95% CI: 0.26 to 1.40, I2= 77.53%, p = 0.01). Variance decomposition showed that within-study heterogeneity was negligible (Tau2 Level 2 = 0.000), while substantial between-study heterogeneity remained (Tau2 Level 3 = 0.419). The detailed results are presented in Figure 5. Figure 6A and 6B presents the consolidated forest plots corresponding to each outcome indicator. Forest plot of overall effects from random-effects three-level meta-analysis results. N = total sample size. Cohen’s d values with 95% confidence intervals (CI), F = F-statistic; I2 = heterogeneity index. Level 2 = variance between effect sizes; Level 3 = variance between studies. Consolidated forest plots illustrating effect sizes for each outcome indicator. Differences in kinetic parameters between patients with ACLR and healthy controls No significant difference in peak external knee extension moment during stance was observed between individuals with ACL reconstruction and healthy controls (k = 8, d = −0.461, 95% CI:−1.086 to 0.164, I2= 73.48%, p = 0.125). Variance decomposition showed that within-study heterogeneity was negligible (Tau2 Level 2 = 0.000), while substantial between-study heterogeneity remained (Tau2 Level 3 = 0.319). No significant difference in peak external knee flexion moment during stance was observed between individuals with ACL reconstruction and healthy controls (k = 4, d = −0.332, 95% CI:−0.945 to, 0.280, I2= 24.38%, p = 0.183). Variance decomposition revealed that within-study heterogeneity was negligible (Tau2Level 2 = 0.000), while only a small amount of between-study heterogeneity remained (Tau2Level 3 = 0.036). The detailed results are presented in Figure 5. Figure 6C and 6D presents the consolidated forest plots corresponding to each outcome indicator. Differences in spatiotemporal parameters between patients with ACLR and healthy controls No significant difference in stride length was observed between individuals with ACL reconstruction and healthy controls (k = 8, d = 0.134, 95% CI:−0.343 to 0.612, I2= 63.49%, p = 0.526). Variance decomposition showed that within-study heterogeneity was negligible (Tau2 Level 2 = 0.000), while substantial between-study heterogeneity remained (Tau2 Level 3 = 0.201). No significant difference in walking speed was observed between individuals with ACL reconstruction and healthy controls (k = 12, d = −0.254, 95% CI: −0.599 to 0.090, I2= 64.07%, p = 0.132). Variance decomposition showed that within-study heterogeneity was negligible (Tau2 Level 2 = 0.000), while substantial between-study heterogeneity remained (Tau2 Level 3 = 0.131). The detailed results are presented in Figure 5. Figure 6E and 6F presents the consolidated forest plots corresponding to each outcome indicator. Effects of time as a moderator Kinematic parameters The meta-regression analysis with time as a moderator showed no significant effect (F = 0.039, p = 0.849). The peak knee flexion angle during stance did not show any meaningful change over time (k = 11, β = 0.003, 95% CI: −0.028,0.033, p = 0.85). Detailed results are shown in Figure 7A. Meta-regression of gait kinematic, kinetic, and spatiotemporal parameters over time. Progressive improvements were observed across all parameters, though global time trends did not reach statistical significance. The meta-regression analysis with time as a moderator showed a significant effect (F = 5.656, p = 0.049). The meta-regression indicated a significant change in peak knee extension angle during stance over postoperative time (k = 9, β= −0.122, 95% CI: −0.243, −0.001, p = 0.049), suggesting a slight improvement over time. Detailed results are shown in Figure 7B. Kinetic parameters The meta-regression analysis with time as a moderator showed no significant effect (F = 0.263, p = 0.626). The peak external knee extension moment during stance did not show any meaningful change over time (k = 8, β = 0.006, 95% CI: −0.021,0.032, p = 0.626). The detailed results are shown in Figure 7C. A total of four effect sizes from independent studies were included in the analysis of peak external knee flexion moment during stance. Due to the limited number of studies and relatively low heterogeneity (I2= 24.38%), meta-regression using postoperative time as a moderator was not conducted. Spatiotemporal parameters The meta-regression analysis with time as a moderator showed no significant effect (F = 0.506, p = 0.503). The stride length did not show any meaningful change over time (k = 8, β = 0.01, 95% CI: −0.026,0.046, p = 0.503). Detailed results are shown in Figure 7D. The meta-regression analysis with time as a moderator showed no significant effect (F = 1.843, p = 0.204). The walking speed did not show any meaningful change over time (k = 12, β = 0.006, 95% CI: −0.004,0.015, p = 0.204). The detailed results are shown in Figure 7E. Discussion This meta-analysis was designed to evaluate whether individuals following ACL reconstruction exhibit normalized gait biomechanics when compared to healthy controls. Among the six key gait parameters examined, only the peak knee extension angle during stance showed a significant difference between individuals following ACL reconstruction and healthy controls. The remaining parameters – including peak knee flexion angle during stance, peak external knee flexion and extension moments during stance, stride length and walking speed – did not differ across groups. These findings indicate that rehabilitation may need to place particular emphasis on restoring knee extension during stance, whereas other aspects of gait may recover more readily with current standard care. Differences in kinematic parameters The current analysis revealed no significant difference in peak knee flexion angle during stance between individuals with ACL reconstruction and healthy controls. This finding is consistent with a systematic review and meta-analysis, which reported no significant differences in stance-phase knee flexion angles during level walking between ACLR individuals and healthy controls [52]. The review also highlighted a trend towards gradual normalization of knee flexion over time, with average peak angles returning to near-normal levels approximately 5 to 6 years post-surgery. Despite the general trend towards normalization, localized joint motion amplitudes may still show differences at certain walking speeds [53,54]. Nevertheless, several earlier studies have reported reduced flexion angles in ACLR patients, often attributing such limitations to postoperative quadriceps weakness or hip abductor deficits that impair sagittal plane control [55]. Additionally, some authors suggest that reduced flexion may serve as a compensatory strategy to decrease anterior tibial shear forces and protect the graft from excessive loading during the early postoperative period [56,57]. The inconsistency across current findings may be attributed to inter-individual variation, differences in rehabilitation protocols, or the inclusion of participants at different postoperative stages. In contrast, a significant difference was found in peak knee extension angle during stance, with ACLR patients showing greater extension than healthy controls. Previous studies have frequently reported joint extension limitations in the early stages after ACL reconstruction, with ACLR individuals showing smaller knee extension angles during stance compared with healthy controls [58]. Compared with these earlier observations, the greater extension angle observed in our analysis may reflect adaptations that occur in later postoperative stages. One possible explanation is that altered gait mechanics and compensatory muscle recruitment patterns over time may contribute to hyperextension tendencies in some ACLR patients [59]. This adaptation could be indicative of an attempt to stabilize the knee joint during late stance in the presence of residual neuromuscular deficits, though further investigation is necessary to clarify this mechanism. Differences in kinetic parameters Our data demonstrate that there were no significant differences in peak external knee flexion or extension moments during stance between individuals with ACL reconstruction and healthy controls. Specifically, no significant between-group difference was observed in peak knee extension moments, a finding that aligns with previous studies reporting that knee extensor torque asymmetries may be present during the early postoperative phase – particularly around 6 months – but tend to resolve by approximately 9 months following ACL reconstruction [60]. This suggests that with appropriate rehabilitation, quadriceps strength and torque control can gradually recover, contributing to restored knee extension mechanics during gait [61]. In contrast, earlier studies have reported reduced knee flexion moments in ACLR patients, often attributing this to proprioceptive impairments resulting from surgical trauma or neuromuscular dysfunction [13]. Such deficits in joint proprioception may disrupt load regulation and lead to more conservative motor strategies, including decreased flexion moments, as a protective mechanism to minimize shear forces on the reconstructed ligament [55]. However, when considered alongside earlier reports of reduced knee flexion moments and neuromuscular or proprioceptive deficits in the early postoperative period [13,55], as well as evidence that quadriceps strength and torque asymmetries can improve with rehabilitation over the first postoperative year [60,61], our pooled finding of no significant between-group difference in knee flexion moments is more likely to reflect the inclusion of participants at different postoperative stages, including some who may have already regained near-symmetric flexion moments, rather than the absence of flexion moment deficits in the early phases after ACL reconstruction. Differences in spatiotemporal parameters The findings demonstrated no significant differences in stride length and walking speed between ACLR patients and healthy controls. Previous research has shown that gait deviations, such as reduced stride length and walking speed in the contralateral limb, are primarily evident at baseline assessments, while evaluations over time or between limbs tend to reveal no significant changes [58]. Although no significant abnormalities were seen in the overall spatiotemporal parameters, alterations in neuromuscular control appeared to persist after surgery [59] and may still trigger more subtle gait changes, including asymmetries or phase-specific modifications in movement patterns [48]. Therefore, careful attention should be directed towards gait symmetry and specific temporal phases to more comprehensively evaluate potential changes in the stride length and walking speed following ACLR. Limitations and future directions A potential limitation of this study lies in the wide range of postoperative time points across the included studies, which introduces potential heterogeneity. It is acknowledged that gait biomechanics are subject to change over time following ACLR, and different follow-up durations may reflect varying stages of recovery. However, the primary objective of this study was not to model recovery trajectories over time, but rather to determine, based on the existing literature, whether patients after ACL reconstruction exhibit gait normalization compared to healthy controls. Second, rehabilitation protocols varied across studies and were often insufficiently reported in terms of duration, frequency and intensity. Such variability may have influenced gait outcomes after ACL reconstruction, yet could not be controlled for in this review due to inconsistent reporting. In addition, sex-specific data were not reported consistently enough to permit stratified analyses, so the potential impact of sex on gait restoration could not be examined. Outcomes such as muscle mass and muscle strength were also rarely reported in a comparable way across studies, precluding pooled analysis of these factors, which are likely to be as important as gait parameters in understanding functional recovery. To address the heterogeneity in follow-up duration, we employed a three-level meta-analytic model that accounts for both within- and between-study variance. In addition, a meta-regression using postoperative duration as a continuous moderator was conducted. This analysis identified a significant time effect only for peak knee extension angle during stance, while the remaining five parameters showed no significant temporal trends. These findings suggest that postoperative duration had limited influence on the overall results, supporting the robustness of our conclusions. Nonetheless, potential residual heterogeneity may still exist. However, insufficient reporting of key variables precluded stratified analyses, limiting our ability to identify potential sources of heterogeneity. Another limitation is the relatively small number of eligible studies, which did not reach the conventional threshold of 10 studies typically recommended for reliable meta-regression. Although meta-regression was conducted to explore heterogeneity, these findings should be interpreted with caution given the restricted study pool. For future research, more standardized longitudinal studies are needed to track gait recovery trajectories at defined postoperative intervals. Such studies should aim to reduce variability in follow-up timing, standardize gait assessment protocols and report key clinical details – including graft type, rehabilitation strategies and return-to-sport status. Conclusion This meta-analysis shows that the majority of gait parameters during stance, including peak knee flexion angle, external knee flexion and extension moments, stride length and walking speed, do not differ significantly between individuals with ACL reconstruction and healthy controls. Only peak knee extension angle remained impaired, though it showed signs of improvement over time. These findings underscore the significance of knee extension control as a key focus for long-term rehabilitation. Supplementary Material PRISMA_checklist.docx Supplementary File1 Search strategy.docx Acknowledgements The authors would like to express their sincere gratitude to all researchers whose work contributed to this review and meta-analysis. Funding Statement This work was supported by the Startup Fund for scientific research, Fujian Medical University (Grant number: 2022QH1315). Disclosure statement No potential conflict of interest was reported by the authors. Data availability statement The datasets analyzed during the current study are available from the corresponding author on reasonable request. References Associated Data Supplementary Materials PRISMA_checklist.docx Supplementary File1 Search strategy.docx Data Availability Statement The datasets analyzed during the current study are available from the corresponding author on reasonable request.