Methods

A systematic review with meta-analysis adhering to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14] was performed. The review was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42022337340).

Literature Search

Six investigators (S.W., P.M., K.B., L.B., C.V.D., and D.F.) identified relevant articles using PubMed, Google Scholar (first 200 entries), Web of Science, and Scopus. The Google Scholar search was conducted in incognito mode to avoid personalized results. The search was performed on 3 April 2023 and updated on 18 August 2025. We used the following string:(case* OR mechanism* OR situation* OR event* OR characteristic* OR movement OR injur* OR scenario* OR rupture* OR torn OR pattern*) AND (ACL OR “anterior cruciate ligament”) AND (video-based OR “video analysis” OR 2D OR 3D OR footage OR televi* OR TV OR recording* OR tapes).

Search results were uploaded to the web interface of rayyan.ai [15]. After removing duplicates, four independent team members screened the titles and abstracts (S.W., P.M., L.B., and C.V.D.). Articles were eligible if (1) published in English or German in peer-reviewed journals, (2) used video analysis to analyze ACL injuries sustained during training or competition, and (3) reported movement patterns performed during the moment of injury. We excluded all articles meeting the following criteria: (1) randomized controlled trial design, case report, or review, (2) analysis of other injuries than ACL rupture, (3) analysis of ACL injuries during nonsporting activities, (4) investigation of nonhuman subjects, (5) in vitro or cadaver studies, (6) execution of a post-injury analysis, and (7) analysis of injury situations by methods other than video recordings.

Eligible studies were discussed regarding inclusion among the review team. Disagreements were resolved through consultation with an additional investigator (D.F.). Full texts were screened using the same inclusion and exclusion criteria. Additional sources were identified through reference lists and a co-citation method using the bibliographic coupling concept (www.connectedpapers.com).

Data Extraction

Five investigators (T.H., D.F., K.B., P.M., and K.H.) independently extracted data from the eligible studies. This included study characteristics (design, source of video recordings, duration, and country of data collection), population details (sex, player experience, performance level, and type of sport), and the frequencies of noncontact, indirect contact, and contact ACL injuries. The data extraction was conducted overall for each study, as individual injury data were not reported in many articles. We used the definition of Luig et al. [16] to categorize the injury mechanism into noncontact, indirect contact, and direct contact.

In several of the included studies, the description of ACL injury events was limited to the executed movement, without further detail on the broader situational context. To account for this, we distinguished between movement patterns (e.g., change of direction) and situational patterns (e.g., ball possession). As a result, a single injury case could be categorized simultaneously under both a situational and a movement pattern. Consequently, overlap between categories was unavoidable, which means that aggregated proportions across categories may not sum precisely to 100%.

For the movement pattern during the moment of injury, we used the following categories: landing (single- and double-leg landing), cutting (i.e., spontaneous, quick change of direction/sidestepping), decelerating, accelerating, ball delivery (kicking, passing, heading), dribbling, tackling/pressing, and being tackled. Tackling/pressing refers to defensive actions to regain possession of the ball, while being tackled refers to an offensive action where the injured player is tackled by a defender. Movement speed was classified as none (standing), low (walking/jogging), or high (running/sprinting) [17–27]. Regarding the game situation, we distinguished between ball possession (yes/no) and tactical situation (offensive/defensive). In addition, the timing of injuries was examined. Injuries were categorized as occurring in the first, second, third, or fourth quarter of a match. For sports without quarter timing (e.g., football, which uses halves), match durations were converted into quarters.

Regarding the participants in the primary studies, we classified their expertise level from 1 (recreational) to 5 (world-class) according to the framework of McKay et al. [28], which utilizes training volume and performance metrics.

We did not consider biomechanical descriptions of ACL injuries in our analysis.

Study Quality Assessment

Risk of bias and methodological quality of the video analysis studies were assessed by three independent investigators (L.R., T.H., and T.G.) using the Quality Appraisal for Sports Injury Video Analysis Studies (QA-SIVAS) scale [11]. The QA-SIVAS scale exhibits high reliability and construct validity for evaluating video analysis studies of musculoskeletal injury. The instrument consists of 18 distinct items, each scoring 0 (no/not stated) or 1 (yes/present). The maximum score is 18, and the quality rating is a percentage value (reached score/maximum score (%)).

Data Synthesis and Statistics

Weighted summary proportions with 95% confidence intervals (CIs) were pooled using meta-analysis of prevalence for the respective movement pattern, movement speed, and game situation variables. We performed separate analyses for (1) all injuries combined, (2) noncontact, (3) indirect contact, and (4) direct contact ACL injuries. As not all included studies consistently reported data using all three subcategories, we also provide pooled data on combinations of direct and indirect, as well as indirect and noncontact injuries, as a supplement (Supplementary Online Material).

Statistical analyses followed established workflows for meta-analysis of prevalence as described by Barendregt et al. [29]. For each study and variable, the frequency of an ACL injury situation or movement pattern was expressed as a proportion (event count divided by the number of injuries with available information for that variable). To obtain summary estimates that are generalizable across studies while acknowledging that underlying “true” proportions may vary between studies owing to clinical and methodological heterogeneity (e.g., sport, playing level, video source/quality, and operational definitions of injury situations), random effects meta-analyses of prevalence were applied. Because inverse-variance pooling of raw proportions can yield unstable variance estimates when proportions approach 0 or 1, study-specific proportions were variance-stabilized using the Freeman–Tukey double arcsine transformation prior to pooling. Between-study variance (_τ_2) was estimated using a DerSimonian–Laird approach and incorporated into inverse-variance random effects weights (1/(v + _τ_2)), such that study weights reflected both within-study sampling variance and between-study heterogeneity. Summary prevalence estimates and corresponding 95% confidence intervals were calculated on the transformed scale and subsequently back-transformed to the proportion scale; statistical heterogeneity was quantified using Cochran’s Q and _I_2 statistics.

In addition to pooled prevalence estimates, moderator analyses were performed using sport type as a factor and proportions as the dependent variable within a random effects meta-regression framework on the transformed scale, applying the same _τ_2-based weighting approach [29]. P values < 0.05 were considered statistically significant. Moderator analyses were conducted only when at least three studies were available. For sports represented by two studies, mean proportions are reported without moderator testing, and for sports represented by a single study, the study-specific proportion is reported descriptively.

Analyses were performed using custom MATLAB scripts (MathWorks, Natick, MA, USA) and jamovi (the jamovi project (2023), jamovi, version 2.3.28).

Equity, Diversity, and Inclusion Statement

Equity, diversity, and inclusion were strategically targeted for the present study. The research and author team consists of female and male researchers with varying nationalities, career levels, professions, and fields of expertise (e.g., exercise science, biomechanics, sports medicine, and trauma surgery). Regarding inclusion criteria for study populations, a diverse group from all geographic locations, sports, and sexes was targeted. However, the study population data included were mostly from high-income countries with no representation of low-income countries.