Introduction

Thigh muscle weakness is common after anterior cruciate ligament reconstruction (ACLR) and increases the risk of reinjury1 and osteoarthritis.2 3 Rehabilitation addressing thigh muscle weakness is the cornerstone of clinical guidelines and recommendations after ACL injury.4–6 While patients are typically expected to regain full quadriceps strength (eg, preinjury capacity, >90% of the uninjured side) within the first year after ACLR to coincide with a return to competitive sport,7 the trajectory of muscle strength deficits within, and beyond, the first year after injury has not been well summarised. With thigh muscle strength often evaluated as a secondary outcome in longitudinal cohort studies and clinical trials, there is an opportunity to synthesise data from hundreds of trials and thousands of participants into meaningful conclusions regarding long-term muscle strength patterns after ACLR.

Several smaller reviews have summarised muscle strength outcomes after ACLR. These studies were limited by measuring single muscle groups only,8–10 focusing on graft comparisons,11 12 including limited strength outcomes (eg, isometric contraction only),8 a lack of quantitative methods13 and a lack of comparison to uninjured controls.14 Two recent reviews exclusively compared ACLR to matched uninjured controls, citing evidence of bilateral deficits in muscle strength as a potential confounder for within-person strength comparisons (contralateral leg).15 16 This is a strength of their analyses; however, it also limited the available pool of studies as within-person comparisons are more common in the literature. There has been considerable debate regarding the value of within-person strength comparisons (ie, symmetry) and whether they may underestimate strength deficits compared with uninjured controls.17–19 Despite recommendations to use normative data for strength comparisons, Limb Symmetry Indices (LSIs) continue to be used by many clinicians20 and are also incorporated in clinical guidelines as part of rehabilitation decision-making.21 In addition to investigating differences in the comparator, understanding variation across contraction types might also be important, given potential differences in muscle output (eg, higher force output in isometric contractions compared with concentric)22 and that clinicians most frequently assess strength isometrically using hand-held dynamometry.23 Analysing results of between-person and within-person comparisons concurrently for the knee extensor and flexors while also separating contraction types has not been reviewed previously.

Few reviews have considered how changing muscle strength and time interact13; for example, a recent review conducted meta-regression with time since surgery,16 but no known reviews have attempted longitudinal analysis with time as a continuous measure allowing for the non-linear effects expected in recovery trajectory. Knowing the expected trajectories of muscle strength may help to guide expectations of patients and clinicians. Therefore, the primary aim of this systematic review was to investigate how knee extensor and flexor muscle strength change over time after ACLR. We aimed to compare muscle strength within person (ie, to the uninjured contralateral limb) and between person (ie, to uninjured healthy populations) to provide the most thorough and robust summary to date. A secondary aim was to investigate the relationship of within-person to between-person muscle strength comparisons.

Methods

This systematic review was reported according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement guidelines.24 It was prospectively registered (PROSPERO: CRD42020216793) as part of a group of systematic reviews investigating changes in lower limb strength and function after ACL injury.25 We made two protocol deviations. The effect measure for our analysis changed from the standardised mean difference to the ratio of means (RoM) for interpretability and to enable the inclusion of studies reporting only LSIs (see the Data analysis section). An additional inclusion criterion was added of publication date 2010 onwards and total n≥50 for feasibility purposes given the number of papers captured (>1100 full-text studies initially eligible). This decision also ensures that results are based on the most contemporary practice available (within last 14 years). We found no evidence of any effect of publication year on outcomes in a sensitivity analysis (online supplemental file 16). All protocol changes were made prior to conducting data analysis.

Eligibility criteria

We included peer-reviewed studies of primary ACL injuries managed with surgical reconstruction, with a mean participant age of 18–40 years. Studies that did not explicitly specify primary injury but reported no revision cases or previous injuries in text (including on the contralateral limb) were included. Studies reporting any revision cases within the population were excluded. We did not restrict based on any other injury (eg, concomitant injuries), rehabilitation (eg, type, duration) or surgical (eg, graft) characteristics or study design.

Studies with a total sample size of ≥50, published in 2010 or later, reporting a quantitative measure (mechanical load, that is, force or moment of force (torque)) of knee extension or flexion (ie, quadriceps or hamstring) muscle strength measured by an isokinetic or hand-held dynamometer were included. Muscle strength had to be reported for the ACL injured limb and for either: (1) the contralateral limb (within-person comparison); and/or (2) uninjured control participants (between-person comparison). For within-person comparisons, we also included studies that only reported results as a LSI. Studies that exclusively reported unilateral strength or mean difference of strength or did not specify the injured side (eg, reported dominant and non-dominant side) were excluded. Studies had to report the timepoint of strength testing to be included, and we included all timepoints up to 10 years post ACLR.

Search strategy

A broad search strategy was employed to capture all research questions in this group of reviews. Search terms were piloted based on key literature and focused on two elements: population (ACL injury) and outcome (strength, physical performance). Free-text and medical subject heading terms were combined and tailored to each database (Medline, Embase, CINAHL, Scopus, Cochrane CENTRAL and SPORTDiscus) (online supplemental file 2). We searched from inception to 28 February 2023, limiting to humans where possible but with no restrictions on language.

Study selection

Two authors (MG and either MHedger or BP) screened titles and abstracts for eligibility, before screening full texts of relevant studies for final inclusion (MG, BM). Disagreements were discussed until consensus, with a third reviewer consulted for any disagreements.26

Data extraction

Data were extracted by one author (MG) and checked by another (JC/MHaberfield). A piloted customised extraction form with data validation was also used to minimise entry errors. Information extracted included study design, country of study, demographics for each group (age, sample size, number of women, body mass index (BMI), graft type), strength measure details, timepoint of data collection (post surgery) and outcome data (reported as injured limb, non-injured limb, control limb (as dominant/non-dominant if reported), or LSI). For longitudinal studies with multiple timepoints, we extracted all available data at all timepoints. Where data were presented only in graphs, we extracted them using Plot Digitizer (V.2.6.8, http://plotdigitizer.sourceforge.net/). Non-English studies were translated using DeepL (http://www.deepl.com).27

Risk-of-bias assessment

We assessed risk of bias on domains outlined by Cochrane Collaboration guidelines (online supplemental file 3): random sequence generation, allocation concealment, blinding of therapist and patient, blinding of assessor, outcome measurement, bias in selection of participants, attrition and statistical analysis.28 Several different study designs were included, and only relevant domains were assessed (eg, bias related to randomisation or group allocation not assessed in cross-sectional studies). Two reviewers (MG and either MHaberfield or BP) assessed the risk of bias independently with any disagreements discussed until consensus, and a third reviewer (AC) consulted if needed. We did not assess the certainty of evidence using the Grade of Recommendations Assessments, Development and Evaluation (GRADE). Most studies in our review were observational in nature, with at least some shortcomings in terms of risk of bias, resulting in automatic downgrades to very-low level certainty evidence. Recognising the studies’ limitations, we took a pragmatic approach and chose not to perform a GRADE assessment, as it was unlikely to provide a meaningful interpretation of the evidence synthesis required to address our research question.

Data analysis

To investigate the change in muscle strength over time, we ran several meta-analyses in RStudio.29 We used the RoM effect size for meta-analyses.30 The RoM expresses the effect as the mean of one group divided by the mean of the comparator. A detailed description on the calculation of variance is published.30 31 The RoM allows for simplified clinical interpretation (eg, a RoM of 0.8 indicates that the ACL limb is 0.8× the strength of the comparator, or 20% deficit), compared with traditional metrics such as the standardised mean difference.30 The RoM is widely used in ecological research and allows for incorporation of studies that present data only as LSIs. The LSI is a mean of ratios, which we approximated to the RoM using formulae (online supplemental file 4). For within-person comparisons (ie, between limbs), the correlation between limbs is accounted for with a slightly modified calculation. Between-limb correlation is rarely reported, so we estimated this based on our own data as r=0.85.32 33

Findings that were not presented as mean (SD) (eg, SE or 95% CI) were transformed with appropriate formulae.28 Data presented as median and IQRs were transformed to mean and SD according to published methods.34 Missing SD data for LSIs were imputed using the ‘simputation’ package using predictive mean matching. We performed single imputation using ‘mean LSI+sample size’ for our matching formula and used all complete cases as donors (781 donor rows). Where studies did not report SDs for within-person comparisons (eg, missing SD for injured/uninjured limb), we calculated an LSI first before using the above approach. We excluded from analyses any groups from studies that used a contralateral knee graft, given the confounding impact this would have on within and between-person estimates. Some outcomes were reported for the same participants across multiple studies. In these cases, we extracted all data before checking for any data duplication and removing as necessary (eg, baseline strength data presented across multiple papers). We classified multiple studies as one ‘cohort’ (ie, the same group of participants across multiple different studies/authors). This allows for the investigation of longitudinal effects even if different timepoints are reported across separate studies. Data from multiple groups per cohort were combined at each timepoint using appropriate formulae (online supplemental file 4).

Data were pooled based on the muscle group (ie, knee extensors or flexors), and the type of comparison: (1) within person (compared with contralateral limb); or (2) between person (compared with uninjured control group). We selected the dominant limb of the control group for comparison where possible. We split our analysis based on contraction types: (1) slow-speed concentric tests (≤120°/s); (2) high-speed concentric tests (>120°/s); (3) isometric tests; and (4) eccentric tests.

We fit a longitudinal mixed-effect meta-analysis with a restricted maximum likelihood estimator, similar to previously reported methods,35 36 using the ‘metafor’ package.37 We used a correlated hierarchical effect model38 to account for situations where a cohort could contribute multiple effects over time, which are correlated. We piloted different working model structures, ultimately fitting random effects for timepoints nested within cohorts, using a continuous autoregressive sample correlation structure. Other options, including further random effects, resulted in overly complex models and/or indistinguishable random effects. We then fit a moderator variable for timepoint, to evaluate the magnitude of muscle strength change over time. We trialled fitting timepoint as a linear, log-linear, polynomial, 3-knot spline and 4-knot spline using model heuristics (lowest Akaike Information Criterion, Bayesian Information Criterion), as well as visual inspection of model fit to determine the most suitable model. Splines were constructed with the ‘rms’ package,39 with knot locations fitted as recommended (knot positions at 0.1, 0.5 and 0.9 percentiles for 3 knots; 0.05, 0.35, 0.65 and 0.95 percentiles for 4 knots).40 We used cluster-robust variance estimation methods in final models (using the ‘clubSandwich’ package41), to provide less-biased SE estimates incorporating the variability and clustering of data present in the model. Findings are presented as the predicted model fit and 95% CIs (for range of timepoints where data were available), along with a prediction interval as a measure of heterogeneity (estimate of where future observations could fall).42 We also calculated a pseudo R2 statistic to estimate the percentage of heterogeneity accounted for by our fixed effect of time point.43

There was insufficient data to run longitudinal analyses for eccentric contraction types, so we fit univariate meta-analyses with a restricted maximum likelihood estimator where three or more homogenous studies were available. We stratified analyses based on time since ACLR as (1) short term (≤12 months); or (2) long term (>12 months) where possible (three or more studies in a group).

An exploratory analysis was conducted to investigate the change in normalised raw strength (Nm/kg) of the ACLR limb over time. We used the same longitudinal meta-analysis approach as above, but with strength in Nm/kg as our effect measure. Only studies reporting their results in Nm/kg units were included. This analysis was performed for slow concentric knee extensor and flexor strength. Results are reported as predicted mean torque in Nm/kg and 95% CIs.

As a secondary exploratory analysis, we used a bivariate model to investigate the relationship between within-person and between-person effects where studies reported both.44 This allowed us to estimate the correlation between the two comparisons, and also to regress the estimated true effects to understand the relationship (slope) between within-person and between-person comparisons (ie, were they representative of each other). Here, each study contributes two correlated effects (within and between person). To account for this, we used a correlated hierarchical effect model fitting random effects for the effect type (within or between person), nested within each study, as well as a moderator variable (fixed effect) for effect type. A regression model was then fit based on the random effects and estimated variance-covariance matrix from the multivariate model, to provide an estimate of the slope.44 45 Because the majority of between-person studies only measured a single timepoint, we used only one timepoint from each study, taking the closest to 12 months post ACLR where multiple were reported. We ran a separate model for quadriceps and hamstring strength, but did not split by contraction type due to data availability. Results are reported with scatter plots, correlation between effect types (rho), as well as the estimated slope relationship between the effect types (online supplemental file 11).

We conducted a sensitivity analysis to investigate the effect of graft site morbidity on strength outcomes by comparing: (1) extensor grafts (bone patellar tendon bone or quads tendon) to other grafts for knee extensor muscle strength; and (2) flexor grafts (any hamstring or gracilis graft) to other grafts for knee flexor muscle strength. Cohorts with mixed graft types were not included in this analysis. We added a ‘graft’ grouping term (eg, extensor graft vs other) as a moderator variable with an interaction term (removed if not significant) to final models from our primary analysis. A second sensitivity analysis was conducted to investigate any differences in isometric strength between isokinetic dynamometers and hand-held dynamometers and by adding a ‘device’ grouping term as a moderator variable to our final models.

Results are presented with plots and the mean difference and 95% CI across all timepoints with available data, with pairwise comparisons constructed using the ‘emmeans’ package.

Publication bias was assessed using a modified Egger regression test using robust variance estimation, accounting for potential dependence of effect sizes.46 Any data that could not be summarised quantitatively were synthesised narratively. Scripts and all data used are publicly available, including all model selection decisions and final models: https://github.com/mgirdwood7/quadham_acl_sr

Equity, diversity and inclusion statement

Our authorship team is gender balanced (4 women, 5 men) with early-career, mid-career and late-career researchers, though all from one country. We have made every attempt to be inclusive of data in this review, including translating four studies from other languages (German, French). We have included data from 29 different countries from Europe, North and South America, Asia, Oceania and the Middle East, though no studies from Africa. Participants included in this review were a mixture of men and women, though some studies reported exclusively on men.

Results

We included 232 studies from 29 countries including 34 220 participants (13 517 women/females, 40%; PRISMA flow chart: online supplemental figure 1.1; study characteristics: online supplemental table 5.1). Most studies compared strength within person (n=207, between person n=44), with a median total ACLR sample size of 80 (IQR 55–125, maximum=4093). Most studies measured strength in the first 12 months post surgery (most frequently measured: 3 months (n=35), 6 months (n=88) and 12 months (n=59)), with 81 studies measuring strength at multiple timepoints (median=2 timepoints, IQR 2–3, max=6). For cohorts with multiple timepoints, the median difference between timepoints was 3.1 months (IQR 3–6 months, minimum 0.7 months, maximum 93.6 months).

Of the 31 171 included ACLR participants, 33% (10 128) received flexor grafts (any type of hamstring graft), 13% (4189) extensor grafts (bone-patellar tendon-bone or quadriceps tendon), 46% (14 340) were in groups with mixed or non-specified grafts and 8% (2514) other graft groups (eg, tibialis anterior). Mean age ranged from 18 to 38 years (median 25, IQR 23–29). Mean BMI was reported by 93 studies, ranging from 21.3 to 28.4 kg/m2 (median 24.2, IQR 23.4–25.0). Isokinetic dynamometers were used for all concentric and eccentric testing, while hand-held dynamometers were used in 10/51 studies (20%) measuring isometric contractions.

For studies with uninjured control groups (44 studies, ACLR total n=2464, uninjured control total n=2833), 14 studies matched control groups on at least one demographic variable, most commonly age (n=12), height (n=6), sex/gender (n=6) and body mass (n=6). Sixty-five per cent (n=29) had similar age, sex and/or BMI (no statistical differences between groups reported).

Risk of bias

Detailed risk of bias summaries are provided in online supplemental figures 6.1 and 6.2. Only 10% of included studies (n=24) were rated as low risk of bias for assessor blinding. Forty-seven studies (20%) had a high risk of bias for outcome measurement—based on not providing replicable detail on methods of strength testing. Selection bias was high or unclear in 32% (n=74) of studies, and more commonly in studies including uninjured controls (78% of between-person studies, n=32). Eighteen studies (8%) were rated as high risk of bias for assessor blinding, outcome assessment and selection and were removed in a sensitivity analysis. The removal of these studies did not alter overall predicted estimates (online supplemental file 13). We found no evidence of considerable publication bias across any of our analyses (online supplemental file 7).

Meta-analyses summary

We present results separated by muscle group and contraction type, with summary plots for each analysis and estimates for 3 months, 6 months, 1 year, 2 years and 5 years in table 1 and online supplemental table 8.2. Non-linear trajectories were most appropriate for all models fit (full details of final models in online supplemental file 8). Overall, muscle strength followed similar trajectories for within-person comparisons, regardless of muscle group or contraction type, with sharp initial postoperative improvement tailing off by approximately 1–1.5 years and minimal improvement after this time. Only slow concentric (≤120°/s) knee extensor and flexor strength had data beyond 5 years post surgery. A lack of data on eccentric contractions was available (11 studies). Knee extensor strength for the ACLR limb was approximately 90% or less of the contralateral limb at 1 year, regardless of contraction type. Knee flexor strength in the ACLR limb was more similar to the contralateral limb for concentric strength, though isometric strength was reduced by more than 10%.

Table 1

Predicted estimates and 95% CI of within-person differences (% of contralateral limb) based on model fit from longitudinal meta-analyses at select timepoints post-anterior cruciate ligament reconstruction (ACLR) for each muscle and contraction type

An exploratory analysis showed knee extensor strength improved by approximately 0.9 Nm/kg from 3-month timepoint to 2.12 Nm/kg at 1 year, peaking at approximately 2 years post surgery. Raw knee flexor strength improved by approximately 0.4 Nm/kg from 3 months to 1.30 Nm/kg at 1 year (peaking at this time).

Between-person comparisons for knee extensor strength at 1 year showed approximately 80% ACLR limb strength compared with uninjured controls. Between-person comparisons for knee flexor strength at 1 year showed no difference to uninjured controls, though were affected by wide CIs and a lack of data.

Knee extensor strengthSlow concentric (≤120°/s)

Within-person comparisons of slow concentric muscle strength were the most common in this review (248 effects from 167 studies, n=23 863, figure 1), showing lower ACLR limb strength to 8.8 years post surgery (table 1). Between-person comparisons (24 effects from 18 studies, n=1223) also showed lower ACLR limb strength to at least 2.4 years post surgery (online supplemental file 8.2). Predicted muscle strength at 1 year was 84.9% of the contralateral limb (95% CI 83.5% to 86.3%) and 80.3% of uninjured control limbs (95% CI 75.1% to 85.9%).

Figure 1

Meta-analysis of within person (left) and between person (right) slow (≤120°/s) concentric quadriceps strength deficits. Red line and shaded region represent the estimated fit (ratio of means, expressed as percentage deficit) and 95% CI, respectively, with grey shading representing the prediction interval. Grey dots represent individual cohorts with black showing linked timepoints across cohorts. n=total sample size, k=number of individual effects.

Fast concentric (>120°/s)

Sixty-seven studies (100 effects, n=7916) compared fast concentric knee extensor strength within person (figure 2), with lower ACLR limb strength at all time points (to 5.1 years, table 1). Eleven studies (15 effects, n=520) evaluated between-person comparisons, with lower ACLR limb strength to at least 4.5 years post surgery (online supplemental table 8.2). Predicted muscle strength at 1 year was 86.4% of the contralateral limb (95% CI 84.4% to 88.5%) and 79.7% of uninjured control limbs (95% CI 73.6% to 86.3%).

Figure 2

Meta-analysis of within person (left) and between person (right) fast (>120°/s) concentric quadriceps strength deficits. Red line and shaded region represent the estimated fit (ratio of means, expressed as percentage deficit) and 95% CI, respectively, with grey shading the prediction interval. Grey dots represent individual cohorts with lines showing linked timepoints across cohorts. n=total sample size, k=number of individual effects.

Isometric

Isometric strength was compared within person by 46 studies (61 effects, n=3964), mostly at 90° knee flexion (43 effects, 70%). ACLR limb strength was lower, to at least 3.3 years (figure 3, table 1). Twenty-seven studies (29 effects, n=1255) measured isometric strength between people, with lower ACLR limb strength at all timepoints (up to 4.5 years, online supplemental table 8.2). Estimated muscle strength at 1 year was 88.8% of the contralateral limb (95% CI 84.5% to 93.3%) and 79.2% of uninjured control limbs (95% CI 73.9% to 85%). A sensitivity analysis showed studies measuring strength at 90° knee flexion had 0.89-fold lower strength deficits (95% CI 0.83 to 0.97) compared with other angles measured (online supplemental file 14). A second sensitivity analysis showed studies measuring isometric strength with an isokinetic dynamometer had 0.91-fold lower strength deficits than hand-held dynamometers (95% CI 0.83 to 0.99, online supplemental file 15).

Figure 3

Meta-analysis of within-person (left) and between-person (right) isometric quadriceps strength deficits. Red line and shaded region represent the estimated fit (ratio of means, expressed as percentage deficit) and 95% CI, respectively, with grey shading representing the prediction interval. Grey dots represent individual cohorts with lines showing linked timepoints across cohorts. n=total sample size, k=number of individual effects.

Eccentric

Three studies measured slow speed eccentric strength within person (three effects, n=237), with results analysed with a univariate meta-analysis (online supplemental figure 10.1). ACLR limb strength was 91.0% of the contralateral limb (95% CI 82.7% to 100.8%). There was insufficient evidence to meta-analyse fast eccentric (one study) or between-person comparisons (one study) (narrative summary in online supplemental file 10.3).

Knee flexor strengthSlow concentric (≤120°/s)

For knee flexor strength, 128 studies (189 effects, n=18 747) measured slow concentric strength within person (figure 4, table 1). The strength of the ACLR limb was lower to at least 8 years. Between-person comparisons from 12 studies (18 effects, n=834) showed lower ACLR limb strength to at least 0.7 years, but large uncertainty due to lack of long-term follow-up studies (online supplemental table 8.2). Predicted ACLR limb strength at 1 year was 93.8% of the contralateral limb (95% CI 92.5% to 95.1%) and 95.5% of uninjured control limbs (95% CI 88.7% to 102.9%).

Figure 4

Meta-analysis of within-person (left) and between-person (right) slow concentric hamstring strength deficits. Red line and shaded region represent the estimated fit (ratio of means, expressed as percentage deficit) and 95% CI, respectively, with grey shading representing the prediction interval. Grey dots represent individual cohorts with lines showing linked timepoints across cohorts. n=total sample size, k=number of individual effects.

Fast concentric (>120°/s)

Fifty-four studies (79 effects, n=5960) measured fast concentric knee flexor strength within person, with deficits to at least 3.6 years (figure 5, table 1). Only 9 studies (13 effects, n=503) compared between people, with major uncertainty affecting estimates (online supplemental table 8.2). Predicted ACLR limb strength at 1 year was 94.0% of the contralateral limb (95% CI 92.0% to 96.0%) and 91.5% of uninjured control limbs (95% CI 79.5% to 105.3%).

Figure 5

Meta-analysis of within-person (left) and between-person (right) fast concentric hamstring strength deficits. Red line and shaded region represent the estimated fit (ratio of means, expressed as percentage deficit) and 95% CI, respectively, with grey shading representing the prediction interval. Grey dots represent individual cohorts with lines showing linked timepoints across cohorts. n=total sample size, k=number of individual effects.

Isometric

Isometric knee flexor strength was compared within person by 27 studies (38 effects, n=1827) (online supplemental figure 9.1), mostly at 90° knee flexion (23 effects, 61%) with deficits in ACLR limb strength shown up to at least 3.4 years (table 1). Twelve studies compared strength between people, with variability in effects resulting in implausible model fitting for later timepoints (online supplemental figure 9.1). Predicted ACLR limb strength at 1 year was 84.5% of the contralateral limb (95% CI 79.7% to 89.6%) and 102.1% (95% CI 89.3% to 116.7%) of uninjured control limbs, though caution is urged with this comparison. A sensitivity analysis showed no effect of joint angle on isometric knee flexor strength (online supplemental file 14). A second sensitivity analysis showed no differences in knee flexor strength deficits when measured with isokinetic or hand-held dynamometers (online supplemental file 15).

Eccentric

Slow eccentric knee flexor strength was measured within person by 9 studies (online supplemental figure 10.2), with univariate meta-analysis showing ACLR limb deficits compared with the contralateral limb of 81.7% (95%CI 74.9% to 89.0%, 6 studies) in the first year post ACLR, and 94.8% (95% CI 87.1% to 103.3%, 3 studies) after 1 year. There was insufficient evidence to meta-analyse fast eccentric (two studies) or between-person comparisons (two studies across slow and fast eccentric), with a narrative summary in online supplemental file 10.3.

Exploratory analysis: change in normalised strength over time

Normalised slow concentric knee extensor strength was reported by 42 studies (56 effects, n=5707). Mean strength rose steadily from 1.27 Nm/kg at 3 months to 2.12 Nm/kg at 1 year, peaking at 2 years (2.32 Nm/kg, table 1, figure 6). Raw slow concentric knee flexor strength was reported in 30 studies (41 effects, n=4603). Mean strength rose from 0.91 Nm/kg at 3 months to 1.30 Nm/kg at 1 year (achieving peak strength, table 1, figure 6). There were limited studies reporting raw strength after the 2-year timepoint (n=5 for knee extensors, n=3 for knee flexors).

Figure 6

Exploratory analysis of normalised ACLR limb concentric for the knee extensor (left) and knee flexors (right). Red line and shaded region represent the estimated fit (ratio of means, expressed as percentage deficit) and 95% CI. Grey dots represent individual cohorts with lines showing linked timepoints across cohorts. n=total sample size, k=number of individual effects.

Relationship of within-person and between-person effects

Exploratory bivariate analyses of studies that reported within-person and between-person comparisons showed that they were not fully equivalent—within-person comparisons underestimated strength deficits relative to between-person comparisons for the quadriceps and hamstrings. The scaling factor based on the model slope was 1.53 (95% CI 1.15 to 2.29) for the knee extensors (ie, between-person deficits were predicted to be 1.53 times greater than within-person ones (online supplemental figure 11.1, k = 27 studies)). For knee flexor strength, uncertainty was much higher suggesting some caution—the scaling factor was 4.57 (95% CI 2.82 to 12.00) (online supplemental figure 11.2, k = 19 studies). The estimated correlations between within-person and between-person comparisons were strong, but not perfect (knee extensor: rho=0.79 (95% CI 0.57 to 0.91), knee flexor: rho=0.66 (95% CI 0.24 to 0.87)).

Sensitivity analyses—effect of graft type

Knee extensor strength was lower in those with extensor grafts (bone-patellar tendon-bone or quadriceps tendon) compared with other grafts for slow concentric (mean ACLR limb strength 0.98-fold lower, 95% CI 0.95 to 1.01), fast concentric (0.93-fold lower, 95% CI 0.91 to 0.95) and isometric knee extensor strength (0.83-fold lower, 95% CI 0.79 to 0.87) (online supplemental figure 12.1). Knee flexor strength was lower in those with flexor grafts (any type of hamstring graft) for slow concentric (0.92-fold, 95% CI 0.91 to 0.93), fast concentric (0.91-fold, 95% CI 0.89 to 0.94) and isometric knee flexor strength (0.84-fold, 95% CI 0.75 to 0.88; online supplemental figure 12.2). Curve trajectories were similar without meaningful differences in shape.

Discussion

This review is the most comprehensive evaluation of knee extensor and flexor muscle strength following ACLR to date. Leveraging data from 232 studies and 34 220 people, we were able synthesise data from concentric, isometric and eccentric contractions across the knee extensors and flexors for a holistic appraisal of muscle strength after ACLR. Our results showed that knee extensor strength at 1 year post surgery remains reduced by more than 10% compared with the contralateral limb, and approximately 20% compared with uninjured controls, with limited recovery after this time—even beyond 5 years post surgery. Knee flexor strength showed smaller deficits (approximately 5%–7% at 1 year compared with the contralateral limb) and had recovered to be similar to the contralateral limb or uninjured controls by 5 years. Graft site morbidity results in 0.85–0.95 times the strength deficit (eg, additional 10% for knee flexor weakness for those with flexor tendon grafts). ACLR limb knee extensor strength deficits measured within person were not equivalent to comparing to an uninjured control limb (between person), with implications for monitoring rehabilitation progress and return-to-sport decision making. Despite robust estimates and CIs, prediction intervals were wide for all analyses, reflecting the complex multifactorial nature of post-ACLR recovery, and the expected variability within clinical populations.

Many people do not achieve common clinical benchmarks of 90% quadriceps strength of the contralateral limb by 1 year post surgery.7 Persistent thigh muscle strength deficits are problematic—they are linked to worse return to sport outcomes,1 higher risk of reinjury,47 and worse long-term joint health (eg, osteoarthritis)48 and knee symptoms.49 Though the quadriceps are often the focus of rehabilitation programmes after ACLR, hamstring strength deficits likely also persist, especially for those who had a flexor tendon graft.50 Based on our findings, a substantial proportion of people after ACLR likely never reach recommended marks of 90% contralateral limb strength, potentially explaining some of the ongoing high burden of knee injury. Within-person deficits appear to plateau after 1–2 years post surgery, when people have usually disengaged from rehabilitation, despite deficits being present. The initial rehabilitation phase appears to be critical for regaining physical capacity, though new trials show that additional exercise programme delivered after this time may be able to ‘top-up’ muscle strength and benefit long-term outcomes.51 52

Normalised ACLR limb knee extensor and flexor strength appear to peak at approximately 2 years and 1 year post surgery, respectively. Unfortunately, there were no longitudinal data following patients more than 2 years after ACLR, limiting our understanding of how raw thigh muscle strength changes in the medium to long terms. The dip in predicted strength seen in our results after the 2-year timepoint should be interpreted with caution, given it is driven only be a handful of cross-sectional studies. Mean peak slow concentric knee extensor strength from our exploratory meta-analysis of 2.3 Nm/kg is well below thresholds of >3.0 Nm/kg suggested previously (~25% deficit).53–55 Our between-person findings for slow concentric strength also confirm this, showing 20% deficit at 1 year post surgery compared with uninjured controls. Slightly lower strength thresholds of approximately 2.5 Nm/kg for knee extensor strength and 1.5 Nm/kg for knee flexor strength have also been proposed,56–58 though our mean estimates are consistently lower than these thresholds across all time points measured. The current implementation of rehabilitation programmes may be inadequate to restore muscle strength of the thigh muscles post-ACLR.57 59

Insufficient rehabilitation could explain some of the muscle weakness seen in our review. Training may not have adequately stimulated hypertrophy and strength gains due to inappropriate prescription, lack of time allocated to recovery and poor adherence. Rehabilitation is a key factor in muscle strength recovery, and something we were not able to assess in this review. Information regarding rehabilitation was not extracted, as it is often inadequately reported (or not at all). We can hypothesise that rehabilitation likely varied considerably, as some studies were clinical trials where rehabilitation was controlled, while others were observational or retrospective. This variability could be seen as a strength—it reflects the variability in clinical practice. As many as 40% of people complete less than 3 months of supervised rehabilitation after ACL surgery,60 and only 2 in 5 people complete more than 9 months.61 Most people do not consult a therapist beyond 12 months post surgery61 suggesting that the duration of rehabilitation may be inadequate for many, as deficits are often still present at this time based on our findings. Clinicians might hypothesise that muscle strength deficits are greatest in those who complete the least rehabilitation; however, we are unaware of empirical evidence to support this notion,62 though better rehabilitation compliance might be linked to better self-reported outcomes and return to sport.63–65 Research is also lacking to support optimal dosage, duration and content of rehabilitation after ACL injury.4 Clinicians should focus on maximising muscle strength gains by ensuring training prescription adheres to evidence-based training principles. However, other barriers to achieving optimal strength after ACLR also need to be considered, including motivation, lack of time, self-efficacy and symptoms.66

Key determinants of muscle force production are muscle size and neural activation, with both likely implicated in the persistent weakness seen after ACLR. Quadriceps and hamstring muscle size (measured as volume or cross-sectional area) is decreased after ACLR and is not restored after rehabilitation is completed.67 68 Graft-specific deficits are also evident, especially after semitendinosus/gracilis grafts where long-term reductions in semitendinosus size result from the surgical disruption of the muscle-tendon unit.67 These findings align with our graft sensitivity analysis, where knee flexor strength deficits were 10%–15% greater in those with hamstring grafts compared with other graft types. Muscle size deficits correlate with the amount of weakness seen post-ACLR but may not fully explain the amount of weakness in force production seen post-ACLR. Central nervous system (CNS) dysfunction may also play a role in persistent thigh muscle weakness and suboptimal rehabilitation outcomes. Alterations to sensorimotor control,69 corticospinal tract structure70 and brain activation are observed following ACL injury.71 72 Deficits in knee extensor strength are associated with corticospinal excitability,73 74 suggesting that persistent weakness is at least partly explained by CNS reorganisation. Hence, even with optimal rehabilitation dosage, thigh muscle weakness may persist if the CNS dysfunction is not addressed, with many authors now calling for implementation of specific training strategies to address this.75–77 High-quality trials of these techniques are the next frontier of ACLR rehabilitation, and an exciting prospect to potentially reduce longer term knee injury burden.

Comparing muscle strength to the contralateral limb is not the same as comparing to an uninjured control limb or population norms.78 Our bivariate analysis was able to synthesise comparisons from the same ACLR samples and showed a discordance of within-person and between-person deficits, with a scaling factor of at least 1.5 for quadriceps strength (ie, deficits were 1.5× larger when compared between person), acknowledging some uncertainty and variability in our estimates. These results suggest that contralateral limb strength is also reduced after ACLR, confirming findings from previous studies.17 18 79 The modest improvement in within-person strength seen in our primary analyses at longer-term timepoints is potentially attributable to the contralateral limb worsening, rather than improving absolute strength of the ACLR limb.18 Contralateral limb deficits may result from reduced activity after injury as well as CNS changes, and potentially manifest shortly after injury (and before reconstruction).80 Bilateral weakness may also have been present prior to ACL injury (ie, potentially a risk factor for ACL injury), though evidence for this is lacking.81

For clinicians, we suggest that even achieving a benchmark goal of 90% contralateral limb strength might underestimate the magnitude of weakness present. Some have suggested a goal of 100% LSI to be a more suitable benchmark.56 82 Comparing normalised torque (Nm/kg) against population norms may be more appropriate,83 though these need to be age and gender appropriate.84 If using contralateral limb strength as a comparator, alternative approaches include measuring strength pre-ACLR (to better estimate a preinjury benchmark),17 or from preinjury strength tests where available. However, these are not without their limitations, as muscle dysfunction could have been present before ACL injury. Unfortunately,