Purpose To compare patient-reported and clinician-reported outcomes, and radiographic results between patients who had had revision ACL reconstruction and those who had had primary ACL reconstruction.
Design Systematic review and meta-analysis
Data sources The MEDLINE, CINAHL, EMBASE and SPORTDiscus electronic databases were searched on 6 August 2015, using 3 main concepts: (1) revision ACL reconstruction, (2) primary ACL reconstruction and (3) treatment outcomes.
Eligibility criteria for selecting studies Articles that compared patient-reported or clinician-reported outcomes or radiographic results between patients who had had revision ACL reconstruction and those who had had primary surgery with a minimum of 2 years follow-up were included. The outcomes evaluated were the Lysholm Knee Scoring Scale, objective International Knee Documentation Committee (IKDC) classification, Tegner Activity Scale, side-to-side difference in anterior tibial translation measured with KT-1000/2000 arthrometer, pivot shift test, tibiofemoral osteoarthritis grading on plain radiographs and subsequent knee surgeries.
Results 8 studies (300 revision ACL reconstructions and 413 primary ACL reconstructions) were included in the meta-analysis. Patients who had had revision surgery reported inferior Lysholm Knee Scoring Scale scores (mean difference: 7.8 points), had inferior clinician-reported knee function as assessed with the objective IKDC classification (IKDC category A: 27% vs 57%; IKDC category C or D: 22% vs 8%) and pivot shift test (grade II or III: 7% vs 2%), and more radiographic evidence of tibiofemoral osteoarthritis (50% vs 25%) compared with patients who had had primary surgery.
Conclusions Revision ACL reconstruction restored similar anterior-posterior knee laxity compared with primary ACL reconstruction. Patients who had had revision surgery reported inferior Lysholm Knee Scoring Scale scores, had inferior clinician-reported knee function and more radiographic signs of tibiofemoral osteoarthritis compared with patients with primary ACL reconstruction.
Statistics from Altmetric.com
Although outcomes after revision surgery have improved as surgical techniques have been refined,3 inferior outcomes are still expected compared with primary ACL reconstruction.4 For example, a recent systematic review that compared the clinical outcomes of revision ACL reconstruction with the Multicenter Orthopaedics Outcomes Network primary ACL reconstruction cohort showed a clinically important difference in the International Knee Documentation Committee (IKDC) subjective knee function score (75 points in the revision surgery cohort compared with 84 points in the primary surgery cohort).3 However, the few controlled trials that have compared outcomes after primary and revision ACL reconstruction,5–12 report conflicting results regarding patient-reported outcomes, return to sport, knee stability and radiographic evidence of knee osteoarthritis.
Therefore, the purpose of this systematic review was to use meta-analysis to compare patient-reported and clinician-reported outcomes and radiographic results from patients who had had revision ACL reconstruction to those who had had primary surgery. Our hypothesis was that revision ACL reconstruction would result in inferior patient-reported and clinician-reported outcomes and worse radiographic results compared with primary ACL reconstruction.
A systematic review with meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analysis guidelines.13
A systematic search of the MEDLINE, EMBASE, CINAHL and SPORTDiscus electronic databases was performed on 6 August 2015. Databases were searched from inception, and search terms were mapped to Medical Subject Headings (MeSH) terms where possible. Search terms were entered under three concepts: concept 1—revision ACL reconstruction; concept 2—primary ACL reconstruction; concept 3—treatment outcomes. Search terms in each concept were grouped with the ‘OR’ operator. The results from each concept were then combined with the ‘AND’ operator to produce the search strategy and final yield (see online supplementary appendix 1 for the search strategy as applied to MEDLINE). The electronic database search was supplemented by hand searching of the reference lists of included articles, and the ePublication lists of leading orthopaedic and sports medicine journals (American Journal of Sports Medicine, British Journal of Sports Medicine, Knee Surgery Sports Traumatology Arthroscopy, Arthroscopy Journal, Journal of Bone and Joint Surgery American and British versions, International Orthopaedics and The Knee).
The following inclusion criteria were applied to the final yield:
Outcomes compared between patients who had had revision ACL reconstruction and patients who had had primary ACL reconstruction;
Age at surgery, sex and time to follow-up reported for revision and primary ACL reconstruction populations;
Minimum 2 years follow-up;
Reported at least one of the following outcomes: Lysholm Knee Scoring Scale score; Tegner Activity Scale score; objective IKDC classification; anterior-posterior knee laxity measured with KT-1000/2000 arthrometer; pivot shift test; radiographic evaluation of tibiofemoral osteoarthritis based on a published grading system;
Published in an English language peer-reviewed journal.
All criteria had to be fulfilled for the article to be included. No restrictions were imposed for graft choice, method of participant matching or cartilage or meniscus treatment. Biomechanical, in vitro and in vivo studies, review articles, surgical techniques, case reports, letters to the Editor, editorials, and conference abstracts were excluded.
To select the articles to be included in the review, first, two authors (AG and GMMM) independently reviewed the title and abstract of each article identified in the literature search. When eligibility was unclear from the title and abstract, the full text of the article was obtained and evaluated for eligibility by the same independent reviewers. Any disagreements were resolved via consensus discussion between the two independent reviewers; a third reviewer was consulted if the disagreement could not be resolved. The same two authors identified the studies to finally include in the meta-analysis, resolving any disagreement via consensus.
Assessment of risk of bias
A modified version14 of Downs and Black's Checklist for the Assessment of Methodological Quality of Randomised and Non-Randomised Studies15 was used to evaluate the risk of bias in included studies. The checklist used in our review comprised 21 items that were rated as ‘fulfilled’ or ‘not fulfilled’. The total number of items fulfilled was summed for each article. Two independent assessors (AG and GMMM) assessed each included article. Consensus discussion was used to resolve any disagreements, and a third reviewer was consulted if the discrepancies could not be resolved. Inter-rater agreement for the risk of bias assessment was evaluated using the κ statistic. The purpose of the assessment of risk of bias was to provide a descriptive summary of the main sources of potential bias in the included studies. Articles were not excluded on the basis of the assessment.
Data extraction and synthesis
Patient demographic details including age at surgery, sex and length of follow-up were extracted to provide an overview of the population. Surgical details including the graft used for primary and revision reconstruction, and meniscal and cartilage status were extracted when reported. The main outcomes extracted for meta-analysis are presented in box 1. Data were extracted and tabulated in an Excel database by one author (AG).
Variables extracted for meta-analysis
Lysholm Knee Scoring Scale
Tegner Activity Scale
Objective International Knee Documentation Committee (IKDC) evaluation
Anterior-posterior side-to-side difference at 134N or manual maximum (measured with KT 1000/2000 arthrometer)
Grading of plain radiograph
Number of subsequent surgeries
The number of patients with knee function classified as normal (IKDC category A), nearly normal (IKDC category B), abnormal (IKDC category C) and severely abnormal (IKDC category D) were extracted. These objective IKDC results were analysed in two ways: (1) the number of knees classified as ‘normal’ (category A); (2) the number of knees classified as ‘normal/nearly normal’ (category A or B) compared with the number of knees classified as ‘abnormal/severely abnormal’ (category C or D).
For knee laxity, the mean side-to-side difference and SD measured in millimetres (mm), and the number of patients with side-to-side difference of (1) <3 mm, (2) 3–5 mm and (3) >5 mm were extracted. Laxity measures were then classified and analysed as the number of patients with side-to-side difference (1) 0–3 mm compared with >3 mm, and (2) 0–5 mm compared with >5 mm.
The number of patients with pivot shift tests classified as normal (negative), nearly normal (1+), abnormal (2+) or severely abnormal (3+) were extracted.16 Pivot shift data were then classified as normal/nearly normal (grades 0 and 1+) or abnormal/severely abnormal (grades 2+ and 3+) for analysis.
Radiographic evaluation of tibiofemoral osteoarthritis was classified according to the grading used in each article.17 Osteoarthritis was defined as: Fairbanks grade ≥1, or IKDC 2000 grade ≥B, or Kellgren and Lawrence ≥2. The number of patients in each group (primary or revision) with radiographic evidence of tibiofemoral osteoarthritis (according to our classification) was analysed.
The number of patients who had had a subsequent surgical procedure involving the index knee, including revision or further revision ACL reconstruction, arthroscopy, hardware removal, meniscal treatment or arthrolysis, was extracted.
The meta-analysis was performed using RevMan V.22.214.171.124 (The Cochrane Collaboration, Copenhagen, Denmark). Continuous variables were extracted and analysed as mean±SD. If the SD was not reported, we contacted the corresponding author and asked him or her to provide the statistic. In the case of no response, the SD was calculated from the available data, according to a validated formula.18 The mean difference (MDiff) and 95% CI were calculated for continuous variables. Relative risk (RR) and 95% CI were calculated for dichotomous variables. We tested for heterogeneity using the χ2 and Higgins’ I2 tests.19
Data were pooled using a random-effects model if statistical heterogeneity was >50% (I2 test); a fixed-effects model was used if statistical heterogeneity was below 50%.20 Demographic (ie, age, sex and time to follow-up) and surgical (ie, graft type and the incidence of meniscal and cartilage lesions) characteristics were combined arithmetically and expressed as means or percentages, as appropriate, for both groups. A p value of <0.05 was considered statistically significant in all analyses.
Following deletion of duplicates, the literature search yielded a total of 616 abstracts (figure 1). A total of 557 abstracts were immediately excluded based on the title and abstract screen; 59 articles were obtained in full text and the selection criteria applied. Fifty-one articles were excluded, as they did not compare outcomes between patients who had had revision and primary ACL reconstruction. Finally, eight studies4–12 were included for meta-analysis (figure 1).
Assessment of risk of bias
Among the included articles, two6 ,7 reported an analysis of prospectively collected data, while the remaining six articles5 ,8–12 were retrospective comparisons. The number of items fulfilled by each article ranged from 11 to 17 (table 1); on average, articles fulfilled 14 out of 21 items (67%). All articles fulfilled the items regarding appropriateness of patient population (items 11 and 12) and description of the surgical procedure (item 4), evaluation (items 2 and 6) and results (items 15 and 16). One article8 performed an a priori power analysis for sample size calculation (item 21; see online supplementary appendix 2). Inter-rater agreement between the two reviewers regarding the assessment of risk of bias was high (κ=0.83 (95% CI 0.74 to 0.92)).
Six articles5–7 ,9–11 matched patients based on age at surgery, sex and time to follow-up. In two articles, the primary surgery group was identified either by randomly selecting patients from a consecutive series (both groups had similar follow-up and age at first trauma),12 or by identifying a consecutive series of patients who had had primary ACL reconstruction (primary surgery group had significantly shorter follow-up compared with revision surgery group).8 Three articles6 ,7 ,10 used a one-to-one matching procedure resulting in equal numbers in both groups; two articles11 ,12 presented similar populations due to a small number lost to follow-up, and one article5 matched a population of primary ACL reconstruction with two groups (of equal size) who had had revision ACL reconstruction (performed with different grafts).
A total of 713 patients were included; 300 had had revision ACL reconstruction and 413 had had primary ACL reconstruction (table 1). Overall, patient demographics were similar between the revision and primary surgery groups. The mean age at surgery was 30.3 years in the revision group and 28.3 years in the primary group. Sixty-three per cent and 59% of patients in the revision and primary groups, respectively, were men. The mean follow-up was 4.5 years in the revision group and 3.9 years in the primary group (table 1).
Revision ACL reconstructions were performed with both autografts (80%) and allografts (20%); the most commonly used were hamstrings tendon autograft (total 43%; ipsilateral 85%, contralateral 15%) and bone-patellar tendon-bone autograft (total 37%; ipsilateral 58%, contralateral 31%, reharvested 11%). All primary ACL reconstructions were performed using autograft tendons: ipsilateral hamstring autograft in 87% and ipsilateral bone-patellar tendon-bone autograft in 13% (table 1).
Seven5–10 ,12 and five6 ,7 ,9 ,10 ,12 articles, respectively, reported the incidence of meniscal lesions in the revision and primary groups (table 1). Medial meniscus injury was observed in 51% of the revision group, lateral meniscus in 27%, and both medial and lateral menisci in 10%. Medial meniscus injury was observed in 41% of the primary group, lateral meniscus in 25%, and both medial and lateral menisci in 3%.
Cartilage status was described in four articles; two articles6 ,7 compared the revision and primary groups. Kartus et al5 reported 54% of patients who had had revision surgery had Outerbridge II–III chondropathy, while Ahn et al8 found chondral lesions in 21% of patients undergoing revision ACL reconstruction. Thomas et al6 found a higher incidence of lateral tibiofemoral and patellofemoral cartilage lesions in patients who had had revision surgery compared with patients who had had primary surgery. Weiler et al7 found the same incidence (20%) of Outerbridge III–IV chondropathy in both groups (table 1).
Lysholm Knee Scoring Scale
Seven articles5–9 ,11 ,12 with 615 patients (251 revision vs 364 primary) reported the Lysholm Knee Scoring Scale score. Patients who had had revision ACL reconstruction reported a significantly lower score compared with patients who had had primary ACL reconstruction (MDiff 7.79 points, 95% CI 6.54 to 9.05 points, p<0.0001; I2=34%, p=0.17; figure 2).
Objective IKDC classification
Six articles5–8 ,10 ,12 with 498 patients (223 revision vs 275 primary) reported the IKDC score results. The pooled percentages for each category were 27% A, 51% B, 18% C and 4% D in the revision group, compared with 57% A, 37% B, 5% C and 0% D in the primary group.5–8 ,10 Based on pooled data, patients who had had primary ACL reconstruction were approximately twice as likely to have knee function classified as ‘normal’ (category A) compared with patients who had had revision ACL reconstruction (RR 1.87, 95% CI 1.20 to 2.89, p=0.005; I2=54%, p=0.05; figure 4A).
When the objective IKDC results were grouped as normal and abnormal, the pooled percentages for ‘normal function’ (categories A and B) and ‘abnormal function’ (categories C and D) were 78% and 22%, respectively, in the revision group, compared with 92% and 8%, respectively, in the primary group.5–8 ,10 ,12 Based on pooled data, patients who had had primary ACL reconstruction were approximately half as likely to have knee function classified as abnormal compared with those who had had revision ACL reconstruction (RR 0.45, 95% CI 0.28 to 0.7, p=0.001; I2=34%, p=0.18; figure 4B).
Anterior-posterior knee laxity side-to-side difference
Eight articles,5–12 with 713 patients (300 revision vs 413 primary) reported mean anterior-posterior knee laxity. When these data were pooled, there was no difference in knee laxity between the revision and primary groups (MDiff 0.18 mm, 95% CI −0.44 to 0.08 mm, p=0.18; I2=0%, p=0.79; figure 5A). Twenty-five per cent of the revision group and 21% of the primary group had greater than 3 mm side-to-side difference in knee laxity;5 ,6 ,8–11 this difference was not statistically significant (RR 0.90, 95% CI 0.67 to 1.21, p=0.50; I2=0%, p=0.82; figure 5B). Six per cent of the revision group and 5% of the revision group had greater than 5 mm side-to-side difference in knee laxity;6 ,8–11 this difference was not statistically significant (RR 1.06, 95% CI 0.50 to 2.24, p=0.88; I2=0%, p=0.80; figure 5C).
Pivot shift test
Five articles6 ,7 ,9 ,11 ,12 with 461 patients (198 revision vs 263 primary) reported the results of the pivot shift test. The pooled percentages were 71% negative test, 24% grade 1+, 5% grade 2+ and 0% grade 3+ in the revision group, compared with 83% negative test, 15% grade 1+, 1% grade 2+ and 1% grade 3+ in the primary ACL group. Based on pooled data, patients who had had primary ACL reconstruction had approximately a quarter of the risk to have an abnormal (grade 2+) or severely abnormal (grade 3+) pivot shift compared with those who had had revision ACL reconstruction (RR 0.23, 95% CI 0.09 to 0.59, p=0.002; I2=0%, p=0.80; figure 6).
Radiographic evaluation of osteoarthritis was reported in three articles with 200 patients (101 revision vs 99 primary).10–12 Fifty per cent of the revision group and 25% of the primary group had abnormal radiographs. Patients who had had primary ACL reconstruction had half the risk of having radiographic evidence of osteoarthritis compared with those who had had revision ACL reconstruction (RR 0.46, 95% CI 0.23 to 0.92, p=0.03; I2=67%, p=0.05; figure 7).
Number of subsequent surgeries
Three articles5–7 reported at least one subsequent surgical procedure in at least one of the revision or primary groups (485 patients; 140 revision vs 345 primary). Subsequent procedures were registered in 11% (n=15) of patients who had had revision ACL reconstruction and in 6% (n=21) of patients who had had primary ACL reconstruction. There was no difference in the number of subsequent surgeries between the two groups (RR 0.60, 95% CI 0.28 to 1.30, p=0.20; I2=49%, p=0.14; figure 8).
Summary of meta-analysis results
Table 2 summarises the patient-reported and clinician-reported outcomes, radiographic results and reoperations reported in each included article. Of the nine parameters evaluated in our meta-analysis, the Lysholm Knee Scoring Scale, objective IKDC graded as category A, objective IKDC graded as category C or D, pivot shift test graded as 2+ or 3+, and radiographic evidence of tibiofemoral osteoarthritis showed significantly inferior results in the revision ACL reconstruction group compared with the primary group. There were no differences in mean anterior-posterior knee laxity or the proportion of patients with >5 mm side-to-side difference in knee laxity, Tegner Activity Scale score, and the number of subsequent surgeries between patients who had had revision ACL reconstruction and those who had had primary surgery.
In this systematic review, we compared the results of controlled studies of patients who had had primary or revision ACL reconstruction. The key findings were that there were inferior patient-reported and clinician-reported outcomes, and worse radiographic results after revision ACL reconstruction compared with primary surgery, despite similar anterior-posterior knee stability.
We found that on average, patients who had had a primary ACL reconstruction reported eight points superior knee function, as assessed with the Lysholm Knee Scoring Scale, compared with patients who had had revision surgery. This difference is similar to a previous study where 491 patients undergoing revision ACL reconstruction were compared with the Multicenter Orthopaedics Outcomes Network primary ACL reconstruction cohort.3 The minimum clinically important difference for the Lysholm score has not been determined, although an eight-point difference likely approaches clinical significance.3 Our results might suggest that patients perceive their knee function is never as good after the ACL has been injured multiple times. This may be an important consideration for rehabilitation clinicians working with patients who have had multiple ACL reconstructions, as it could influence the collaborative functional goals that clinicians and patients set. It also highlights the importance of clinicians helping patients establish realistic expectations of what their knee function might be after surgery before they choose to have a revision ACL reconstruction.
In contrast to the Lysholm Knee Scoring Scale results, the physical activity level at follow-up (as measured with the Tegner Activity Scale) was similar between the two groups. However, it is important to consider the lack of clear differentiation between the preinjury and preoperative physical activity level and the fact that the Tegner Activity Scale also includes occupational function so may not be ideal for discriminating functional level. Therefore, the Tegner Activity Scale may not be the best outcome measure for evaluating return to physical activity. These factors could weaken the conclusion that return to an acceptable level of physical activity (including participation in sport) is possible after revision ACL reconstruction. On the other hand, one study9 included in our meta-analysis compared return to sport rates between patients who had had primary ACL reconstruction and those who had had revision surgery, and found no difference between the groups. Younger patients may be more likely to return to sport following ACL reconstruction,21 although the articles included in our meta-analysis had comparable study populations, particularly regarding age and sex, which could have minimised the risk of selection bias influencing the physical activity participation results.
While it appears that revision ACL reconstruction restored anterior-posterior knee stability similarly compared with primary ACL reconstruction, dynamic rotational laxity (as measured with the pivot-shift test) was not fully restored after revision surgery. There was an almost fourfold increased risk of residual rotational laxity in the revision ACL reconstruction group compared with the primary surgery group. We offer three possible explanations for this. First, a repeatedly injured knee could have undiagnosed lesions of the secondary restraints that are believed to be important for rotational stability and pivot shift biomechanics (including the anterolateral ligament or lateral capsular ligament).22–25 In support of this is the fact that good clinical results and rotational stability have been achieved when adding extra-articular lateral plasty or tenodesis to an intra-articular reconstruction.26–28 Second, a higher incidence of lateral meniscal lesions or defects after revision ACL reconstruction compared with after primary surgery might contribute to an increased pivot shift after revision ACL reconstruction.29 Third, hardware presence, tunnel enlargement or previous tunnel malposition may present technical challenges for the surgeon performing a revision procedure. It is possible that these factors could result in suboptimal femoral and tibial tunnel placement; and in turn, a vertical femoral tunnel could affect the control of rotational laxity.11 Residual rotational laxity combined with the higher incidence of meniscal and cartilage lesions found during revision ACL reconstruction (compared with primary surgery) could explain the inferior clinician-reported results despite adequate anterior-posterior stability.
When the global status of the knee is considered, longer time since the original injury, and the primary and secondary trauma itself, coupled with repeated insults to the knee (eg, surgery, tunnel drilling30) and release of inflammatory mediators12 could help to explain the almost twofold increased risk of tibiofemoral osteoarthritis after revision ACL reconstruction. The extent of cartilage injury at the time of surgery may also play a role.17 Patients undergoing revision ACL reconstruction have already had at least one previous serious knee injury (and surgery), and this could increase the risk of osteoarthritic changes. In support of this, previous reports of long-term radiographic outcomes after primary ACL reconstruction5 show superior results after both isolated and combined ACL injury compared with the results of revision ACL reconstruction reported in our review. Returning to sport (particularly pivoting sports) is likely to be accompanied by a higher risk of reinjury to the ACL31 and other knee structures including the meniscus and articular cartilage. Therefore, given the history of knee injury and the risk of future knee injury, it raises the question: should the athlete who has had revision ACL reconstruction even return to pivoting sports? This is a difficult question to answer, but shared decision-making, where the athlete understands the reasonable options and risks based on information that is presented without bias, may help him or her make an informed decision.32
The clinician-reported evaluation of knee status measured with the objective IKDC score also showed an almost a twofold increased risk of clinical failure (knee function classified as abnormal or severely abnormal) after revision ACL reconstruction. The objective IKDC classification is a composite score that evaluates multiple aspects of knee function (range of motion, joint stability, harvest site pathology and joint crepitus). The suboptimal results after revision ACL reconstruction could be due, not only to these signs and symptoms, but also to the repeated surgeries with multiple graft harvesting. This may be especially the case with bone-patellar tendon-bone grafts, which were used in 37% of revision surgeries, but in only 13% of primary reconstructions, and have been associated with increased risk of anterior knee pain.33
Finally, regarding the reoperation rate, the small numbers of patients evaluated do not allow us to draw strong conclusions, although it could be argued that with its low reoperation rate, revision ACL reconstruction may represent a relatively safe procedure.
Our meta-analysis has some important limitations. First, by definition, all the patients included in the revision group had already had at least two ACL reconstructions on the index knee, and had already undergone the procedure used as the control comparison (ie, primary ACL reconstruction). However, the attempt to match patients’ demographic characteristics in both groups strengthens the likelihood that we were comparing two groups that only differed on one variable—primary or revision surgery. On the other hand, the matching method was not consistent between the studies, and important variables such as meniscal and cartilage status have not been considered in the matching process. Second, while not a limitation of the review, per se, there are few high-quality studies comparing outcomes between patients who have had primary ACL reconstruction and those who have had revision surgery. The relatively small number of studies included in our review limited the data pooling that was possible. For example, there were insufficient data for meta-analysis of other clinical scores (such as the IKDC subjective knee evaluation or Knee Injury and Osteoarthritis Outcome score) that might capture the knee status and sport participation more accurately. Third, given that the minimally clinically important difference for the Lysholm score is unknown,34 it is possible that the MDiff of eight points may not be clinically significant.
Finally, the lack of a common evaluation of radiographic outcomes and the need for hybrid grading also limits our ability to make strong conclusions regarding osteoarthritis.
Anterior-posterior knee laxity was restored in a similar manner after revision ACL reconstruction compared with primary ACL reconstruction. However, patients who had revision surgery reported significantly inferior Lysholm Knee Scoring Scale scores, had inferior clinician-reported knee function as assessed with objective IKDC and pivot shift test, and more radiographic signs of tibiofemoral osteoarthritis compared with patients having their first ACL reconstruction.
How might it impact on clinical practice in the future?
Patients perceived their knee function was not as good after the ACL had been injured multiple times. Therefore, it is important for clinicians to help patients establish realistic expectations of what their knee function might be after surgery, before the patient chooses to have revision ACL reconstruction.
What are the findings?
After revision ACL reconstruction:
Anteroposterior knee stability was restored similarly to primary ACL reconstruction, while rotational stability was not.
Patient-reported outcomes were inferior compared with primary ACL reconstruction.
There were worse radiographic results with more degenerative changes.
Twitter Follow Clare Ardern at @clare_ardern
Contributors AG and GMMM performed the search and performed the analysis; SZ was third reviewer and wrote the manuscript; MPN helped with search, tables, figures and manuscript management; CLA performed the search, reviewed manuscript and helped with clinical relevance and bias assessment; MM critically analysed the manuscript
Competing interests None declared.
Patient consent Obtained.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.