Exercise-Based Training Strategies to Reduce the Incidence or Mitigate the Risk Factors of Anterior Cruciate Ligament Injury in Adult Football (Soccer) Players: A Systematic Review

Anterior cruciate ligament (ACL) is one of the most concerning injuries for football players. The aim of this review is to investigate the effects of exercise-based interventions targeting at reducing ACL injury rate or mitigating risk factors of ACL injury in adult football players. Following PRISMA guidelines, a systematic search was conducted in CINAHL, Cochrane Library, PubMed, Scopus, SPORTDiscus and Web of Science. Studies assessing the effect of exercise-based interventions in ACL injury incidence or modifiable risk factors in adult football players were included. 29 studies evaluating 4502 male and 1589 female players were included (15 RCT, 8 NRCT, 6 single-arm): 14 included warm-up, 7 resistance training, 4 mixed training, 3 balance, 1 core stability and 1 technique modification interventions. 6 out of 29 studies investigated the effect of interventions on ACL injury incidence, while the remaining 23 investigated their effect on risk factors. Only 21% and 13% studies evaluating risk of injury variables reported reliability measures and/or smallest worthwhile change data. Warm-up, core stability, balance and technique modification appear effective and feasible interventions to be included in football teams. However, the use of more ecologically valid tests and individually tailored interventions targeting specific ACL injury mechanisms are required.


Introduction
Football (soccer) is one of the most popular sports, with more than 260 million players around the world [1]. Football is also a sport exposed to a high risk of injury, considering that the overall injury incidence is 6.6 injures per 1000 players hours [2]. Given that being exposed to a high number of injuries reduce the chances to sporting success [3], injury management (i.e., mitigation and maximising player availability) is one of the most concerning issues in football clubs. Specifically, lower injury incidence rates has been correlated to superior performance (i.e., higher league position, more games won, more goals scored, greater goal difference and total points) in professional football [4], while injuries that cause a high injury burden (i.e., those requiring a high number of days lost, such us ligament sprains and joint injuries to the knee and the ankle) are more likely to impact negatively on team performance [5]. At the team level, the performance decrement associated to a high injury incidence can lead to losses of~£45 million per season, on average, in English Premier League teams [6].
For the football player, one of the most concerning injuries is the anterior cruciate ligament (ACL) injury given its devastating consequences, such us the increased risk of developing early posttraumatic knee osteoarthritis [7,8], or the high rate of reinjuries to the graft or the opposite knee [9][10][11][12][13]. Furthermore, in professional football, only 60% of players who ruptured their ACL competed at the highest level 5 years later [14]. Additionally, ACL injuries in football causes, on average, an injury burden of 29.8 days per 1000 h of exposure [15]. Even though the ligament injury rate in European male professional football have decreased during the 2000s [16], the ACL injury rate showed no declining trend [17]. Therefore, efforts aimed at reducing the rate of ACL injury in football appear to require further development, being non-contact or indirect contact injuries the specific focus, as they correspond to the 88% of all ACL injuries [18] and could be targeted by injury risk mitigation programs [19]. Consequently, these non-contact or indirect contact ACL injuries will be the focus of this systematic review.
ACL injury is not an injury with a high incidence, with an average of 0.43-0.60 injuries per team per season in football professional teams [17,20]. As a result of the low number of ACL injuries suffered per season, some potentially effective interventions could not be able to reach significance when rate is used to evaluate their efficacy [21,22]. As an alternative, in the purpose of reducing ACL injuries, intervention programs could be also developed with the aim of modifying risk factors of ACL injury [23], and efficacy could be assessed by measuring changes in proxy factors (i.e., surrogates of injury) [24,25]. Specifically, the identification of modifiable risk factors would increase the potential for screening athletes at higher risk and targeting interventions to address the specific mechanisms that increase ACL injury risk [26,27]. In football, ACL injuries mostly occur during cutting actions [18,28,29] with visual observational analyses confirming a mechanism of knee valgus, abducted hip, flat and externally rotated foot, and ipsilateral trunk tilt and contralateral rotation [18]. These aberrant movements have been shown to increase the multiplanar knee joint loads and, hence, increase the ACL load [30,31]. Additionally, neuromuscular factors have been shown to have implications for ACL injury [23,27]. In this sense, different metrics of hamstrings-to-quadriceps (H/Q) ratio could potentially provide useful information regarding the load that the ACL is assuming, as coactivation of hamstrings and quadriceps could protect the knee against anterior shear forces at the tibia [32]. Furthermore, balance ability could have an influence in ACL injuries, given the positive contribution of a higher hamstrings, hip and trunk muscle activation in supporting and reducing knee joint loadings [33]. Although findings are still inconclusive, movement quality and competency deficits may be further linked to greater joint loads [26]. Other risk factors such as restricted ankle or hip mobility could predispose the knee higher loads and increase the risk of ACL injury [34,35]. Therefore, some of these biomechanical, neuromuscular and physical capabilities could be potentially targeted with the purpose of modifying the risk factors of ACL injury [36].
To date, several systematic reviews have been published in the field of ACL injury prevention in athletes [21,23,[37][38][39][40][41]. The last research in football players is the work published by Grimm and collaborators [21], where only randomized-controlled trials (RCT) in which ACL injury incidence was reported were included. Even though RCTs are considered level 1 of evidence, they are usually difficult to implement in the "realworld" environment, and other study designs (i.e., non-randomized, NRCT) could be additional sources of evidence used by practitioners to guide their practice and provide recommendations to their athletes [42]. On the other hand, at the best of our knowledge, the most recent systematic review that took into consideration the effect of preventive programs in both ACL injury incidence and risk factors of ACL injury in football players was conducted 12 years ago, by Alentorn-Geli et al. [23]. However, given that football has evolved [43] and new strategies to screen injuries have been developed in recent years [44], Table 1. Criteria for inclusion according to the PICOS method.

Definition Inclusion Criteria Exclusion Criteria
Population Adult (≥16 and ≤40 years old) football players (i.e., Association football) of any level Both male and female Not having suffered a severe injury the previous 2 years Studies including different cohorts of athletes apart from football players (e.g., basketball, volleyball, handball) in which no sub-analysis by sport was performed

Intervention
Exercise or training-based interventions lasted at least 4 weeks, performed twice a week Interventions performed with exogenous modalities (i.e., bracing, taping, etc.) or those exercise-based interventions lasting less than 4 weeks. Comparator Control group data if available (although not necessarily) No exclusion criteria by comparator Outcome Either contact or non-contact ACL injury incidence or rate of injury Test measurements evaluating any modifiable risk factor (i.e., potentially targeted by exercise-based interventions) previously reported to have an influence in ACL injury [23,26,27,46], such as biomechanical, neuromuscular and/or physical tests (e.g., biomechanics of landing or cutting actions, H/Q ratio, balance measures, ankle or hip range of motion, etc.) Overall injury incidence not explicitly reporting ACL type injuries Test measurements evaluating non-modifiable risk factors (e.g., anatomical) Study design

Randomised-controlled trials
Non-randomised studies Single-arm studies Systematic reviews, meta-analysis, conference papers, book chapters or studies published in languages other than English.

Study Selection
Two authors (JOJ and AFR) independently performed the study selection based on title and abstract screening. Full-text of those studies in which there were not absolute evidence for their exclusion, were retrieved and further analysed. Those cases where a discrepancy existed were resolved by an in-depth discussion between the authors (JOJ and AFR). If disagreements persisted, they were solved by a third author (TDS). PRISMA flow diagram for the description of the overall process is depicted in Figure 1.
tests (e.g., biomechanics of landing or cutting actions, H/Q ratio, balance measures, ankle or hip range of motion, etc.)

Study design
Randomised-controlled trials Non-randomised studies Single-arm studies Systematic reviews, meta-analysis, conference papers, book chapters or studies published in languages other than English.

Study Selection
Two authors (JOJ and AFR) independently performed the study selection based on title and abstract screening. Full-text of those studies in which there were not absolute evidence for their exclusion, were retrieved and further analysed. Those cases where a discrepancy existed were resolved by an in-depth discussion between the authors (JOJ and AFR). If disagreements persisted, they were solved by a third author (TDS). PRISMA flow diagram for the description of the overall process is depicted in Figure 1.

Data Extraction
Data of the included articles were subsequently extracted in an Excel spreadsheet in the same way that literature search was performed. The following data was extracted from each study: title, author(s), publication year, participant sex, participant age, number of participants in experimental and control groups, participant level, compliance rate, number of dropouts, supervisor of the program, type, contents and characteristics of the intervention, comparison group, outcome of interest, reliability data, smallest worthwhile change and changes in the outcome measures of interest from baseline to following the intervention. If the case of data elements of interest were missing or unclear, the study authors were contacted for clarification. In those studies in which no effect sizes (ES) were

Data Extraction
Data of the included articles were subsequently extracted in an Excel spreadsheet in the same way that literature search was performed. The following data was extracted from each study: title, author(s), publication year, participant sex, participant age, number of participants in experimental and control groups, participant level, compliance rate, number of dropouts, supervisor of the program, type, contents and characteristics of the intervention, comparison group, outcome of interest, reliability data, smallest worthwhile change and changes in the outcome measures of interest from baseline to following the intervention. If the case of data elements of interest were missing or unclear, the study authors were contacted for clarification. In those studies in which no effect sizes (ES) were reported, the ES were manually calculated (ESc) from the extracted data using the corrected Hedges'g proposed by Turner et al. [47]. The scale proposed by Hopkins et al. [48] was used for interpretations of the magnitude of results, whereby the magnitude of ES was considered as trivial (≤0.20), small (0.20-0.59), moderate (0. 60-1.19), large (1.20-1.99), or very large (≥2.00).

Risk of Bias Assessment
Risk of bias for RCTs was assessed using the Version 2 of the Cochrane Tool for assessing risk of bias in randomised trials (RoB 2) [49]. In the case of NRCTs, the tool used was the Risk of Bias in Non-Randomized Studies of Interventions tool (ROBINS-I) [50].
The above tools for assessing risk of bias are domain-based evaluation tools that are currently the frequently and preferred method to assess the credibility of study findings over quality checklists and quality scales [51]. The RoB 2 is a domain-based risk of bias assessment tool rigorously structured into five bias domains: (1) bias arising from the randomisation process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in measurement of the outcome, and (5) bias in selection of the reported result [49]. Since the aim of the review was to investigate the effect of assignment to the intervention at baseline instead of the effect of adhering to it (i.e., effectiveness vs efficacy), an intention-to-treat analysis was considered. For risk of bias assessment of cluster-RCT, a supplement of the RoB 2 proposed by the same authors was used. Depending on the judgments in each of them, an overall risk of bias for the result is established. The ROBINS-I has been designed in order that confounding factors and co-interventions that may potentially lead to bias in NRCT could be identified [50]. Previous injury, level, age, sex and exposure to training and competitions were considered confounders as they can influence the risk factors of ACL injury [23,27].
Two authors independently performed the RoB 2 and ROBINS-I (JAN and JMS) tools. After agreements and disagreements were discussed, in cases in which there persisted any disagreement, a third author (JOJ) was consulted for the final decision.

Study Selection
Details of the search results are represented in Figure 1. The systematic search performed through the six previously mentioned databases provided 4394 records which were gathered in Endnote X9 software. After duplicates were identified and removed, 2813 records remained, reducing to 102 following screening by title and abstract. After full-texts were retrieved and further analysed, 29 studies met the eligibility criteria and were included in the systematic review. One study was excluded because, although the proposed programme consisted of 3 weeks of preventative training plus 5 months of maintenance, no information was reported regarding the maintenance phase [52]. Therefore, it is not possible to determine if this part targets the same components than the previous phase, and then, the programme did not reach the minimum 4-week duration established as reference criteria. Reasons for exclusion are detailed in Figure 1. No study of the references lists met the eligibility criteria.

Characteristics of Included Articles
The 29 studies included in the systematic review investigated the effects of exercisebased interventions on either reducing the incidence of ACL injury (6 studies) [53][54][55][56][57][58] or mitigating risk factors of ACL injury (23 studies) , where a sum of 6091 adult (male: 4502, female: 1589) football players were evaluated.
Finally, the 6 single-arm studies with no CG investigated the effect of RT-based [71,74], MT-based [70,76] and WU-based interventions [61,77] on H/Q ratio [61,71,74], different measures of balance ability [77], or biomechanics of dynamic tasks [70,76] of ACL injury, with no study examining its effect on ACL injury incidence ( Table 4). None of the singlearm studies reported directly measured reliability data or the SWC. Two studies reported compliance rate [61,70], and two reported the supervisor of the programme [70,76]. In Figure 2, a depiction of the different exercise-based interventions used to target both ACL injury risk mitigation and ACL injury reduction, for each study design, is presented.

Risk of Bias in Individual Studies
A summary of the risk of bias in both parallel and cluster RCT at the different domains level of the RoB 2 tool are displayed in Supplementary Table S1 and Figure S1. Only one study was reported to be at low overall risk of bias [73], while 4 present some concerns [62,65,78,80], and 10 are at high risk of bias [53,[56][57][58]60,63,64,68,69,81]. Regarding the 8 NRCT, 6 of them were classified as critical risk of bias [54,55,59,67,72,79], while two were classified at moderate risk of bias [66,75] (Supplementary Table S2 and Figure S2). Single-arm studies will be only discussed as potentially feasible implementations whose effectiveness should be further investigated. Therefore, no risk of bias analysis was conducted. A detailed explanation of this section is included in the Electronic Supplementary Material.

Results of Individual Studies
Among the 29 included studies in the review, 6 studies evaluated the effect of exercisebased programs on ACL injury incidence. Four studies were RCT, in which three warm-upbased interventions (i.e., PEP Program [53], and FIFA 11+ [56,57]) found a protective effect on ACL injury incidence, while one in which part 2 of FIFA 11+ was performed at the end of the training showed no differences against the traditional FIFA 11+ [58]. Two studies were NRCT, one including FIFA 11+ [54] and the other a MT-based interventions [55], both of which demonstrated no protective effects on ACL injury incidence.
23 studies evaluated the effectiveness of exercise-based programs on different risk factors of ACL injury. 13 studies were RCT. Six studies used warm-up-based interventions (i.e., FIFA 11+), four of them showing positive results in improving different metrics of static and dynamic balance [60,63,73], GRF asymmetries in countermovement jumps and conventional H/Q ratios at 1.05 and 3.14 rad/s [65], and five showing no effect on balance [60,73], ankle and hip ROM, asymmetries in hop tests [60], frontal plane angles during a DVJ [80], conventional H/Q ratios at 60 and 240 deg/s [80] and Functional Movement Score (FMS) [78]. Two balance-based interventions showed positive results on different metrics of balance [68,69]. Two RT-based interventions found positive [64], and no [62] improvements in different metrics of H/Q ratios. One MT-based intervention [81] was effective at improving some metrics of the FMS. 6 additional studies were NRCT. One warm-up intervention did not improve drop-jump biomechanics [59]. Three RT-based interventions were effective in improving H/Q ratio [67,72,79], while one was effective at reducing bilateral between-limb differences in isokinetic knee extensors and flexors strength [67], and other was not [79]. One balance-based intervention was effective at improving balance metrics [75]. One TM-based intervention was effective in improving COD movement quality and performance [66].

Risk of Bias in Individual Studies
A summary of the risk of bias in both parallel and cluster RCT at the different domains level of the RoB 2 tool are displayed in Supplementary Table S1 and Figure S1. Only one study was reported to be at low overall risk of bias [73], while 4 present some concerns [62,65,78,80], and 10 are at high risk of bias [53,[56][57][58]60,63,64,68,69,81]. Regarding the 8 NRCT, 6 of them were classified as critical risk of bias [54,55,59,67,72,79], while two were classified at moderate risk of bias [66,75] (Supplementary Table S2 and Figure S2). Singlearm studies will be only discussed as potentially feasible implementations whose effectiveness should be further investigated. Therefore, no risk of bias analysis was conducted. A detailed explanation of this section is included in the Electronic Supplementary Material.

Results of Individual Studies
Among the 29 included studies in the review, 6 studies evaluated the effect of exercise-based programs on ACL injury incidence. Four studies were RCT, in which three warm-up-based interventions (i.e., PEP Program [53], and FIFA 11+ [56,57]) found a protective effect on ACL injury incidence, while one in which part 2 of FIFA 11+ was performed at the end of the training showed no differences against the traditional FIFA 11+ [58]. Two studies were NRCT, one including FIFA 11+ [54] and the other a MT-based interventions [55], both of which demonstrated no protective effects on ACL injury incidence. 6 studies were single-arm studies. Two used warm-up-based interventions that were effective at improving H/Q ratios [61] and balance [77]. Two used RT-based interventions, effective at improving functional, but not conventional H/Q ratio [71,74]. Two used MBbased interventions, effective at improving stop-jump task biomechanics [70,76].
When analyzing the findings of individual studies according to their risk of bias (Supplementary Tables S1 and S2), the only study at low risk of bias found that FIFA 11+ was effective at improving time-to-stabilization during a single-leg landing task (−1.8%), but not at improving core stability or SEBT [73]. Two studies at moderate risk of bias also found no effect of FIFA [80] and FIFA 11+ [78] on H/Q ratio and knee valgus angle during a countermovement jump, and in FMS, respectively. Contrasting to these findings, other high-risk studies investigating the FIFA 11+ found positive effects of the programme on both ACL injury incidence [54,57] and several risk factors of injury [59,60,63] (Tables 3 and 4). Other studies at moderate risk of bias found positive effects of RT-based interventions on functional H/Q ratio during fatigued conditions [64] but not in conventional, unfatigued H/Q ratio [62]. Dello Iacono et al. [65] found a positive influence of a core stability training on conventional H/Q ratio and in GRF asymmetries during one-leg countermovement jumps (CMJ). Finally, the study carried out by Dos'Santos et al. [66] showed promising results, as their TM-based programme was effective at improving performance and movement quality during COD. The rest of studies were considered to be at high or critical risk of bias. Therefore, their findings must be considered with caution.

Discussion
To our best knowledge, this is the first systematic review which investigates the effects of exercise-based interventions on ACL injury incidence and risk factors in football players from that of Alentorn-Geli et al. [23] published 12 years ago. The summary of findings and their implications will be summarised in the following sections: (Section 4.1) Effects of exercise-based interventions on ACL injury incidence; (Section 4.2) Effects of exercisebased interventions on risk factors of ACL injury; and (Section 4.3) Potentially feasible interventions from single-arm studies. Following this, a discussion of the importance of different risk factors and practical applications will be provided.

Evidence from Randomised-Controlled Trials
Four cluster-RCTs investigated the effects of warm-up-based interventions on ACL injury incidence showing varied results ( Table 1), all of which were classified as high risk of bias (Supplementary Table S1 and Figure S1). Gilchrist et al. [53] and Silvers-Granelli et al. [56,57] found that PEP Program and FIFA 11+, respectively, could effectively reduce the ACL injury incidence. In the case of FIFA 11+ there was a 46.1% reduction of injuries in the IG, with 70 number-needed-to-treat (i.e., 70 athletes needed to be treated to prevent 1 ACL injury). Even though a similar reduction (i.e., 43.8%) in injury rates was found in female athletes who participated in neuromuscular training (FIFA 11+ could be considered neuromuscular training) [82], number-needed-to-treat data should not be compared since it is time-dependent [83] and different designs have been used for its calculation (i.e., RCT of one competitive season vs. systematic review with different time windows studies) [56,84]. These findings are in line with a systematic review concluding that FIFA 11+ is effective at reducing injuries in football players [84]. In spite of this, surprisingly only 10% of FIFA's member governing bodies associations have implemented the programme with researchers highlighting a low compliance with the 11+ [85,86]. This low adoption and compliance could be explained, among others, by the time required to complete (~20-25 min), boredom associated with the programme [85], and the soreness and fatigue caused by exercises (i.e., Nordic curls) contained in Part 2 [85,86]. To prevent these issues and improve compliance, Whalan et al. [58] proposed a rescheduled FIFA 11+, in which parts 1 and 3 were performed during the warm-up, while part 2 was performed at the end of the training. However, no differences against traditional FIFA 11+ were observed. Given that effectiveness partly relies on the compliance and implementation of any prevention programme [87], it remains unknown if FIFA 11+ could effectively reduce ACL injury rate in the "real world", particularly at the sub-elite and community level.

Evidence from Non-Randomised Trials
The two NRCT investigating the effect of exercise-based intervention on ACL injury rate [54,55], both classified as critical risk of bias, did not demonstrate any reduction in ACL injury rate (Table 2), even though one MT-based programme was effective at reducing severe knee injuries [55]. The fact that this reduction was achieved within a short duration MT-based protocol per session (i.e., 12 min approx.) warrants further investigation through higher quality NRCT or RCT. In Table 5, an overview of the findings is provided for the effects of the exercise-based interventions on ACL injury incidence (Section 4.1).  Table 1). The most frequently warm-up administered was the FIFA 11+. Some studies reported that the FIFA 11+ was effective at improving different surrogates of ACL injury risk [60,63], while others found no differences compared to a CG [63,73,78,80]. Notably, the only RCT in the present systematic review assessed at low risk of bias found no meaningful differences between the FIFA 11+ and CG in SEBT following a 9-week intervention, although an improvement, albeit small, in time-to-stabilisation in IG (−1.8%) was observed [73]. However, the practical implication of such small change into overall ACL risk of injury remains unknown. Although the FIFA 11+ has been shown to decrease ACL injury rate, the controversy of the findings (some studies reporting positive while others no effects for a given component) makes difficult to elucidate the mechanisms by which the programme may be effective at decreasing the incidence [87]. The effectiveness of a warm-up-based stability training in both H/Q ratio and asymmetry in GRF during CMJ [65] requires further research to corroborate these findings. Additionally, of note, changes in H/Q ratio should be considered with caution, as they can be obtained through different changes in the hamstrings and quadriceps musculature (i.e., numerator and dominatory; for example, a reduction in quadriceps strength could improve the ratio, but probably not the risk of injury), thus the constituent components of any ratio should always be examined. Table 5. Summary of evidence regarding the efficacy of exercise-based interventions on ACL injury incidence and risk factors of ACL injury coming from randomized-controlled trials and nonrandomized studies.

Injury Incidence Risk Factors of ACL Injury
• FIFA 11+ [56,57] and PEP Program [53] appears to be effective at reducing ACL injury incidence. • FIFA 11+ still present some pitfalls that restricts its implementation in football teams due to duration, boredom and soreness issues [85,86].

•
The programme proposed by Krutsch et al. [55] appears to be an effective and feasible option to decrease severe knee injuries, but should be further investigated through a RCT specifically investigating ACL incidence.

•
The high sample size or time needed to permit a high ACL injury occurrence which satisfies statistical power is extremely difficult to achieve [53], and thus, complicating the development of rigorous intervention studies investigating their effect in ACL injury incidence.
• FIFA 11+ may be effective by eliciting positive adaptations in terms of balance [60,65] or dynamic stabilization [73], but not in mobility, asymmetry in hop tests [60], fundamental movement patterns [78] or biomechanics of jump landing tasks [59] in adult football players.

•
The core stability programme proposed by Dello Iacono et al. [65] seems to be effective at either improving H/Q ratio and GRF asymmetry during CMJ.

•
The balance programme proposed by Gioftsidou et al. [68] may be effective at eliciting moderate to large improvements in balance.

•
The TM program proposed by Dos'Santos et al [66] appears to be effective at improving both performance and movement quality of cutting actions (main mechanism of ACL injuries [18]). • All of the above interventions appear to provide the best effectiveness/feasibility balance to be implemented in the real soccer context, although they should be further explored through low risk of bias RCT designs.
ACL anterior cruciate ligament, CMJ countermovement jump, GRF ground reaction force, H/Q hamstrings to quadriceps, RCT randomized controlled trial, TM technique modification.
Two studies included a balance-based intervention in order to improve balance and stability [68,69], both showing a positive effect. Balance training has been proposed to be an effective component in both reduction of ACL injury rate and mitigating the predisposing factors [27,88]. Recently, balance training has been considered an effective strategy to mitigate risk factors of ACL injury during COD [89], potentially attributable to positive changes in hamstring, hip and trunk muscle activation, which supports and reduces knee joint loads [33]. However, the high frequency of training needed (i.e., 6 times a week) [68] and the small magnitudes of the effects [69] question the effectiveness and feasibility of these interventions. Additionally, the absence of clear differences between unstable and stable surfaces [69] in balance are in line with previous works [90], although the potential reductions in strength and power application as a consequence of training on unstable surfaces should be considered when they are implemented [91].
A 7-week RT-based intervention showed promising results improving functional H/Q ratio in both fatigued and unfatigued states [64]. RT has been commonly proposed to be a critical component of successful injury prevention programmes in football and is central for facilitating positive tissue adaptations and robustness [92]. Among its potentially beneficial effects include improved coordination, enhanced technique in different activities, reduction of potentially hazardous joint loads, and improvement of the psychological perception of high-risk situations seem to be possible mechanism for reducing the risk of suffering acute injuries, such us ACL tear [93]. However, the requirements of gym-based equipment and the high weekly volume (i.e., >20 min performed 3/week) [64] makes difficult its inclusion into most football training contexts.
Of note, it must be highlighted that only three out of nine studies directly measured the reliability of their data [65,68,78], while one acknowledges other reliability data [60,64] and four studies did not report any reliability data at all [63,69,73,80]. Furthermore, only one study [73] established the SWC. Taken collectively, the findings of the studies must be interpreted with caution, because it is uncertain whether the training-induced changes exceeded the measurement error, and therefore casts doubt whether they were "true" or "real" findings.

Evidence from Non-Randomized Trials
The most common intervention investigated in a NRCT design were RT-based interventions, showing effectiveness at improving H/Q ratio and bilateral differences in peak torque flexor/extensor ratios [67,72,79], in line with the previously mentioned advantages that RT possesses in relation to decreasing the risk of injury [92,93]. However, the serious methodological issues present in their designs resulted in critical risk of bias classification.
A balance-based intervention, similarly to which was found in RCT, found positive effects of a football-specific balance programme on static balance assessments [75], which aligns with the literature evidence suggesting that balance training as a component [87], and specificity as a feature of prevention programmes [94] may positively influence in the risk of injury.
Two studies evaluated the effectiveness of their interventions on lower-body biomechanics of dynamic tasks that try simulate the mechanisms of ACL injury (i.e., landing/COD) [59,66], showing no meaningful effect of FIFA 11+ on peak knee flexion angle during a drop vertical jump [59] and an improvement of a TM-based intervention on COD performance and quality of movement [66]. The findings of the latter intervention are appealing because of several factors: (i) it is one of the only two NRCT studies at moderate risk of bias; (ii) it only requires two 20-min sessions a week which can easily implemented on the field; (iii) both performance and injury risk are improved which is likely to improve athlete and coach adherence and compliance; and (iv) the results were practically meaningful (compared against SWC and SDD). In the absence of robust risk factors of ACL injury [22], it seems to be worthwhile developing training modalities that aim at reducing potentially high-risk postures associated with ACL injury mechanisms [95], where COD has been widely shown to be the main injury mechanism [18,20,29], and which are effective both from a performance and injury perspectives, as they will be more easily adopted by coaches and practitioners [31].
Similar to the aforementioned RCTs, it must be highlighted that only 2 out of 6 studies reported reliability and SWC data [59,66], while the other 4 failed to did not report any information related to reliability measures [67,72,75,79]. Therefore, again, it is difficult to determine the practical relevance of the results reported. In Table 5, an overview of the findings for the effects of the exercise-based interventions on risk factors of ACL injury is provided.

Potentially Feasible Interventions from Single-Arm Studies
Single-arm study characteristics are displayed in Table 3, although due to the previously mentioned limitations as a consequence of not containing a CG, interventions which could be easily implemented in any context (i.e., no sophisticated equipment and so much time required) will only be discussed. Some of the single-arm studies proposed interventions required machine and free-weights [70,71,76] which is equipment not easily available in any training facilities, or they correspond more to full training programmes (i.e., 60-min RT sessions twice a week, plus 60-min field conditioning session twice a week) than preventative interventions [70,76]. Regarding only those potentially feasible in the field, two preventative programs effective at improving muscle activity (i.e., part 2 of FIFA 11+ [77]) or H/Q ratio (i.e., RT-based [74]) may merit further research through multiplearm, RCT. However, until then, their possible influence on ACL injury risk remains, at best, speculative.

Do All Risk Factors Established Elsewhere in The Literature Equally Contribute to ACL Injury?
Many risk factors have been proposed to be related to ACL injury throughout the years [27,30,[96][97][98][99][100][101]. Among them, it is modifiable risk factors which have risen in interest given that they can be targeted by preventative programmes [102]. Neuromuscular (e.g., antagonist-agonist relationships, muscle activation, decreased co-contraction, decreased proprioception) and biomechanical (i.e., ankle, knee, hip and trunk movements in three planes of motion) deficits have been frequently proposed as main modifiable risk injuries of ACL injury [23,96,99], also specifically in football players [23]. Although some of these factors have been proposed to be related ACL injury [103,104], others have presented opposing results (e.g., no relationship between FMS and injury) [105].
With the final purpose of reducing the likelihood of suffering an ACL injury, screening tests should be validated, in a three-step process [22]: (i) a strong relationship between a marker from a screening test and injury risk must be found in prospective studies, (ii) the test properties of the marker must be validated in relevant populations, with appropriate statistic techniques, and (iii) an intervention programme targeted to athletes at high risk must be more beneficial that the same program targeted to all the athletes. Nevertheless, even these prospective studies also have limitations, such us the fact of relating the determined marker with an injury that occurs weeks or months later [106].
In football players, ACL injury mechanisms have been widely established [18,28,29,107]. In the two most recent visual observational analysis studies of ACL injury mechanisms in professional players (male and female) [18,28], it was concluded that performing a COD while pressing or tackling was the main common situation, where ipsilateral trunk tilt and contralateral rotation, abducted hip, dynamic knee valgus and flat and externally rotated foot were the most frequently observed mechanisms [18,28]. Other studies have also found COD and landings to be the main non-contact ACL injury mechanisms in football [29,96] and American football [107]. Therefore, in the absence of strong evidenced risk factors to be targeted, evaluating movement quality and identifying biomechanical and neuromuscular deficits during potentially high-risk maneuvers (i.e., landings and CODs) can provide important information regarding an athlete's "injury risk profile" [108,109].
Three-dimensional motion analysis is the gold standard for evaluating biomechanical variables; however, they require high-cost equipment and are time-consuming to be applicable in football [44]. To overcome these limitations, several cost-effective qualitative field-based screening tools have been developed, such us the landing error scoring system (LESS) [110], tuck jump assessment (TJA) [111], qualitative analysis of single leg loading (QASLS) [112], CMAS [109] or the 2D video analysis scoring system of 90 • CODs [113] to evaluate risk of ACL injury. Since these assessments have been developed to identify biomechanical and neuromuscular control deficits similar to the direct mechanisms of ACL injury (i.e., landings, cuttings), and they can be simply evaluated in the field context, future research is necessary that investigates the effect of exercise-based training interventions on biomechanical and neuromuscular control deficits (movement quality) during landing and cutting field-based screening tests in footballers [22]. Furthermore, even though a better movement quality in common mechanisms of ACL injury could mitigate the risk of injury, few published prevention programs appear to expose athletes to the task-specific elements of the injury mechanisms [95], a field that warrants further research.

Practical Applications
Based on the findings of the systematic review, warm-up-based interventions such us FIFA 11+ or PEP Program seem to be effective at reducing ACL injury rates [53,56,57].
These aforementioned field-based methods can be easily performed at the community to elite level in football because they do not require sophisticated equipment, and generally take only 20 min. However, the mechanisms by which these interventions are effective remains unanswered from the findings of the present work, although from studies with preadolescent female football players, it can be speculated that it is effective at reducing knee valgus angles and moments during landings, probably due to the large volume dedicated to jump-landing technique [114,115]. Nevertheless, it appears to be a poor modality at inducing favorable changes in cutting technique, most likely to the low volume of cutting drills in the programme [114,115]. In the case of FIFA 11+, though, it presents some pitfalls which potentially limits its implementation in football teams despite its effectiveness in reducing injury rates. Additionally, it could be more appropriate and specific in relatively inexperienced players, particular at the community level, as this could be a sufficient training stimulus to elicit improvements in balance, strength, or NM control. Some core stability [65], balance [68], and TM interventions [66] have shown promising results, since they can potentially provide protective benefits in a feasible way (i.e., basic equipment, and no more than 20-min sessions, 5/week interventions). Conversely, some of the included RT-based interventions have been shown to be effective at targeting neuromuscular factors, although the interventions used makes difficult their translation to the real context.
In order to reduce ACL injury rates, it is necessary that the proposed interventions are both effective and feasible (i.e., providing protective effects and easily implementable in the real context), and not only show efficacy in ideal conditions [116]. This can be achieved by developing shorter and more flexible prevention programmess that facilitates its inclusion in football teams [89]. There is a scarcity of RCT and NRCT investigating the effect of exercise-based interventions on movement quality tasks simulating the main mechanisms of ACL injury (i.e., landing and COD), an area which warrants further research. Additionally, none of the interventions have been individually designed to target the identified specific high-risk factors of the football player (e.g., knee valgus, poor trunk control). Although general interventions would be more easily implementable, the more feasible individualization in elite teams also warrants further research. Finally, despite the effectiveness of mixed multicomponent program as a way to reduce ACL injuries [117], none of the included interventions included the main components in the way in which have been previously recommended [27,118]. The use of these multicomponent programs, which could potentially improve also other physical qualities, would improve the external validity of these interventions.

Limitations
Our systematic review is not free of limitations. Firstly, studies including team sports athletes other than football players were excluded from the analysis. However, it could be argued that some interventions could be useful for different team sports. Given the uniquely characteristics of the different sports, including football, the aim of the present systematic review was to gather only those specifically carried out in football players. Secondly, a meta-analysis was not conducted, although a high number of studies were included. The two main reasons were: (i) there were three types of different study designs which should not be mixed; and (ii) within each type of study designs, interventions, comparators and outcomes of interests were too heterogeneous to conduct a proper metaanalysis. Additionally, given the low number of ACL injuries per team per season, most of the study interventions aiming at reducing ACL injury incidence found difficulties to reach significance, hence masking their potentially protective effects. Although expanding the literature search to overall knee injuries would solve this problem, the variability of injuries, with their associated specific risk factors, would make it hard to provide specific recommendations. Finally, when analysing ratios as potential outcomes for injury risk mitigation (i.e., H/Q ratio), it should be considered that they may not change because of both components increasing (i.e., strengthening of knee flexors/extensors) or decreasing (i.e., weakening), which would drastically determine the interpretation of the change. However, these sub-components analyses are not always performed in research investigations.

Conclusions
Our findings revealed that some exercise-based strategies could be potentially effective at both reducing ACL injury rate and mitigating risk factors of ACL injury in adult footballers. Warm-up-based interventions such us FIFA 11+ or PEP Program appear effective and feasible in the context of football teams across community to elite levels, although the mechanisms by which they are effective are not well understood. Core stability, balance and TM-based interventions appear to be better options to include as preventative programmes, as they are feasible and meaningfully decreased risk factors of ACL injury. Other included RT-based interventions have shown protective effects in neuromuscular deficits related to ACL injury (i.e., H/Q ratio), although their time and equipment required makes difficult to potentially be implemented in football teams. Consequently, for future work, improvements in risk of bias frequently observed in RCT and NRCT is required, as well as further research into the effect of potentially effective and feasible interventions in movement quality in relation to sport-specific task similar to the main mechanisms of ACL injury, such as landing and cutting tasks. More research analyzing the effect of interventions targeting individual risk factors of ACL injury are also needed. Finally, considerations for future research are discussed in Table 6. Table 6. Summary of considerations and recommendations for future research.

Developing preventative frameworks focusing on movement quality in risky movements
Of the 29 studies included, only four evaluated the effect of any exercise-based intervention on movement quality during potentially risky movements associated with the common ACL injury mechanism in football [59,66,70,76], while only one study has been carried out with a cost-effective tool which could be easily implemented in the football context [66]. Even though the effect of several interventions on movement quality has been performed with different team sports including football players [37], the specific nature of some exercise-based adaptations and the uniquely context in which football occur [119] could justify the need of developing more football-specific preventative programs. Furthermore, only one NRCT study incorporated TM as a strategy to mitigate risk factors of ACL injury [66] despite it is widely known the contribution of biomechanical determinants in increasing ACL loads [120], and knowing its promising results in other sports [121] and in young female players [122]. On the other hand, given the influence that neurocognition may have in ACL injuries [123], open skills tasks that evaluate athletes under football-specific neurocognitive demands (e.g., unanticipated COD) should be included in screening tests assessing effectiveness of intervention programs [123].

Improving quality of the interventions
None of the included studies had pre-registered the study protocol before to its execution, by which it may be speculated that this is not common trend in Sports Sciences. Since this pre-registration would allow to compare evaluation and data analysis finally carried out with those initially intended, it would be easier to detect risk of bias, especially that related to bias in the se-lection of reported outcomes [49]. Although problems arise from the inability of blinding athletes and care providers are sometimes unavoidable in the context of interventions carried out in football teams, others such us bias due to confounding variables in NRCT or bias due to deviations from intended interventions and unequal training volumes in RCT can be prevented with appropriate analysis performed (i.e., pre-registration and overall transparency with the research process) [50]. Randomization and concealment of allocation sequence processes should be improved and explicitly reported. By doing this, the number of low risks of bias studies (only one in the present review) would be higher and, therefore, findings more reliable [49]. Additionally, by increasing samples and/or the follow-up times, greater statistical power would be reached in the associations, especially those with ACL injury incidence [21]. Table 6. Cont.

Recommendation Rationale
Appropriate reporting of important features of the program: reliability of outcomes, SWC, compliance and supervisor It is also suggested that reliability and SWC data are directly measured so that practical relevance of the results obtained because of an exercise-based intervention could be determined [90], and to ensure that training induced changes exceed the measurement error to increase the certainty improvements are "real". This is extremely pertinent when researchers do not have the opportunity to utilize a CG and therefore adopt a single-arm design. In the present review, only 5/24 and 3/24 of the studies evaluating risk of injury variables have reported directly measured reliability and SWC data, respectively. Of note, it is suggested that supervisor of the interventions is specified, as it is known the potential positive influence of the quality of the feedback provided (i.e., through verbal, auditory and visual cues) [124][125][126] in the reinforcement of proper technique during anterior cruciate ligament injury prevention exercises [127]. 17/29 included studies reported who were the intervention supervisors. Additionally, despite the positive relationship that has been shown between compliance and effectiveness of exercise-based interventions targeting injury reduction [57,58], only 8/29 studies reported compliance rates. Indeed, compliance to the intervention has been shown to be a critical component of prevention programmes, as it highly determines its effectiveness [87]. Therefore, going forward, before concluding a training modality as potentially ineffective, it is central to consider the training compliance which, unfortunately, 72% of studies in this review failed to report. Thus, it is suggested to incorporating such data in future research to confirm the efficacy of injury mitigation training interventions.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/. Risk of bias in individual studies, Table S1: Risk of bias in different domains of randomised controlled trials (i.e., RoB 2), Table S2: Risk of bias in different domains of non-randomised trials (i.e., ROBINS-I), Figure S1: Risk of bias graph of the randomised-controlled trials (RCT) (i.e., Cochrane's RoB 2), Figure S2 Informed Consent Statement: Not applicable.

Data Availability Statement:
The data within this systematic review and is available through the relevant articles referenced throughout.