Studies were eligible if they included aerobic gymnasts (any sex, age group, or competitive level) and evaluated associations between modifiable physical fitness characteristics and/or training-related factors and injury outcomes. Eligible physical fitness exposures included, but were not limited to, neuromuscular performance and movement qualities such as strength, power, strength-endurance, balance, flexibility or range of motion, motor control, and body composition or anthropometrics when analyzed as potentially modifiable correlates. Eligible training-related exposures included measures of training exposure, frequency, volume, intensity, internal and/or external training load, training monotony/strain, periodization characteristics, and competition density or congested schedules when examined in relation to injury outcomes.

The primary outcome domain was injury occurrence in aerobic gymnasts, operationalized as any reported injury incidence, prevalence, rate, risk, or time-to-event, including acute and overuse conditions, regardless of whether the study used a time-loss, medical-attention, or any-complaint definition. For clarity, we treated risk as the probability of a new injury over a defined follow-up (prospective designs), and correlates as cross-sectional or retrospective associations with injury history or current problems (22). We also conceptualized pain and musculoskeletal problems as potentially overlapping with injury outcomes, consistent with health-problem surveillance frameworks that capture time-loss and non–time-loss conditions. Studies were required to report an injury outcome quantitatively and to present an association between at least one eligible exposure and injury, or sufficient data to derive an association. Observational designs (prospective or retrospective cohort, case–control, and cross-sectional studies) were eligible. Controlled trials were eligible only if they reported injury outcomes and provided analyzable associations between baseline or training-related exposures and subsequent injury. Case reports, case series with very small samples, narrative reviews, systematic reviews, conference abstracts without sufficient methodological detail, editorials, and opinion pieces were excluded.

No restrictions were applied to publication year. Reports were considered regardless of language. Where a full text required translation, translation was undertaken using automated translation with verification by a professional fluent in the language where feasible.

PubMed, Scopus, and Web of Science Core Collection were searched on 26 January 2026, and this date was recorded as the last search date for each source. In addition, the reference lists of all included studies and of relevant review articles identified during screening were examined to identify additional eligible studies.

Search strategies were developed to capture three two core concepts: aerobic gymnastics (including common synonyms such as sport aerobics and competitive aerobics), and injury outcomes. The PubMed strategy used a combination of keywords and, where applicable, controlled vocabulary terms; analogous term sets were adapted for Scopus and Web of Science Core Collection to reflect platform-specific syntax and field tags. No date limits were applied, and language limits were not used. The specific search strategy is shown in Table 1.

All records retrieved from the searches were exported to Zotero for de-duplication, after which the deduplicated library was uploaded to a systematic review screening platform for study selection. The author screened titles and abstracts against the eligibility criteria, with an independent, blinded external expert screening in parallel. Both screened all records at title/abstract stage and all retrieved full texts in duplicate, independently and blinded to each other’s decisions. Inter-rater agreement was quantified as percent agreement, showing 93%. Full texts were retrieved for all records deemed potentially eligible or unclear, and the author and the blinded external expert independently assessed full texts for inclusion. Decisions were cross-checked, and discrepancies at either stage were resolved through discussion and consensus; when agreement could not be reached, a second independent blinded external expert adjudicated. When information required to determine eligibility was missing or ambiguously reported, attempts were made to clarify eligibility by contacting study authors.

Data were sought for injury outcomes at the broadest level reported by each study, including overall injury occurrence and, where available, stratified outcomes such as acute versus overuse injury, anatomical region-specific injury, time-loss injury, and injury severity metrics. When studies reported multiple compatible operationalisations of injury (for example, both medical-attention and time-loss outcomes), all eligible outcomes were extracted and clearly mapped to their definitions and observation windows. If multiple time points were reported, the earliest time point reflecting prospective injury occurrence during follow-up was prioritized for primary extraction, while additional time points were retained for contextual interpretation.

Data were extracted on participant characteristics (sample size, sex distribution, age, competitive level, training history, and growth/maturation indicators when available), study and season context, and methodological features relevant to internal and external validity. Data were extracted on exposure measurement methods (test protocols for fitness variables, monitoring methods for training variables, timing of exposure assessment relative to injury surveillance, and units of measurement). Data were also collected on injury surveillance procedures (injury definition, method of ascertainment, reporter type, and exposure-time denominator when used). Data extraction was performed by the author and independently verified by a blinded external expert using a extraction form. Discrepancies were resolved by consensus with reference to the full text.

We evaluated internal validity for each included study with the Quality in Prognosis Studies (QUIPS) instrument (26), which is designed for reviews examining prognostic or risk-factor relationships between an exposure and subsequent outcomes. The author and an independent, blinded external expert appraised all studies in parallel, rating the likelihood of bias within each QUIPS domain: participant selection, loss to follow-up, measurement of the prognostic factor, measurement of outcomes, handling of confounding, and the adequacy of statistical methods and reporting. Domain ratings were then combined into an overall risk-of-bias classification using predefined decision criteria that placed particular weight on identification and adjustment for key confounders and the validity and reliability of measurements for discipline-relevant exposures (e.g., training volume, competition density, internal load indices, and biomechanical/alignment measures) and injury outcomes. Because several included studies were cross-sectional or retrospective survey designs, we applied QUIPS using design-adapted decision rules: (i) for study attrition, one-time surveys were rated based on participation/non-response bias and completeness of key fields (rather than loss to follow-up), and (ii) for prognostic factor measurement, we rated the reliability/validity of retrospective exposure measurement and the risk of differential recall (particularly when exposures and outcomes were self-reported over long windows). Assessments were cross-checked, and any discrepancies were resolved through discussion to reach consensus, with adjudication by a second independent blinded expert when necessary.

For each exposure–injury association, effect estimates were extracted as reported, including odds ratios, risk ratios, hazard ratios, incidence rate ratios, regression coefficients, correlation coefficients, or other appropriate measures, along with corresponding precision estimates such as confidence intervals, standard errors, or p-values. Where necessary for interpretability across studies, effect estimates were harmonized in direction so that the referent category and the direction of higher risk were explicitly stated. A priori selection rules were applied when multiple eligible estimates were available. We prioritized adjusted models over unadjusted models when the covariate set included at least age/sex and prior injury or exposure; then we prioritized time-loss outcomes over any-complaint outcomes when both were available because of greater clinical specificity, while retaining any-complaint outcomes for burden interpretation; and finally when multiple operationalisations of the same construct were reported, we prioritized the measure with the clearest definition and greatest sport-specific relevance.

Studies were first organized into prespecified synthesis groupings based on the primary exposure domain, separating physical fitness factors from training-related factors, and then further grouped by specific construct (for example, strength/power, flexibility, balance/motor control, or training load/exposure). Within each grouping, studies were also classified by design (prospective versus retrospective/cross-sectional) and by injury outcome definition (for example, any injury versus time-loss injury, and acute versus overuse where available) to support meaningful interpretation.

A structured narrative synthesis was undertaken, supported by detailed evidence tables that presented study characteristics, exposure operationalisation, injury definitions, analytical models, and extracted effect estimates with precision. When multiple studies evaluated conceptually similar exposures and comparable injury outcomes, consistency of direction and magnitude of associations was examined qualitatively, with explicit attention to study design and risk of bias when interpreting patterns. Quantitative pooling was considered inappropriate when fewer than three studies assessed a comparable exposure–outcome pair, when injury definitions were not compatible, when exposure units/operationalisations could not be harmonized, or when risk of bias was predominantly high in outcome measurement/confounding domains.

To prepare data for synthesis, units and operational definitions of exposures were recorded verbatim, and where studies reported multiple operationalisations of the same construct, the most sport-relevant or most clearly defined measure was prioritized for the primary narrative while retaining alternative measures for sensitivity of interpretation.

Figure 1 presents the PRISMA 2020 flow diagram summarizing the study selection process. Database searching identified 3,721 records (PubMed, n = 749; Scopus, n = 1,886; Web of Science, n = 1,086). After removal of 761 duplicates, 2,960 records were screened by title and abstract, and 2,938 were excluded at this stage. Full texts were sought for 22 reports. Ten reports were excluded following full-text assessment due to no injury outcomes (n = 8), no specific population (n = 1), or no injury–exposure analysis (n = 1). In total, 12 studies met the inclusion criteria and were included in the systematic review.

Table 2 summarizes the methodological characteristics of the included studies, presenting the design features.

Table 3 summarizes the risk-of-bias appraisal of the 12 included studies using the QUIPS tool across the six standard domains (study participation, study attrition, prognostic factor measurement, outcome measurement, confounding, and statistical analysis/reporting), together with the overall judgement for each study to support interpretation of the strength of the available evidence.

Table 4 synthesizes studies examining athlete-related screening and prediction factors for injury in aerobic gymnastics–related populations. This table is organized around evidence on physical and biomechanical characteristics and risk-prediction approaches, including regression-based screening models and emerging machine-learning/early-warning systems. It includes studies that primarily evaluated individual-level attributes and/or algorithms intended to classify injury risk or discriminate injured versus uninjured gymnasts.

Table 5 summarizes studies reporting injury burden and training- or context-related correlates, including epidemiological descriptions of injury/pain prevalence, anatomical distribution and severity, as well as associations with training exposure, competition density, technical training context, and selected psychosocial or prevention/healthcare factors. This table captures the descriptive and explanatory evidence base most directly aligned with exposure and contextual determinants of injury risk and burden in aerobic gymnastics.

Injury burden and clinical presentation were consistently high where adult competitive aerobic gymnasts were sampled, and lower, but still meaningful, in youth aerobic gymnasts. In world-level aerobic gymnasts, 72.3% reported at least one injury in the prior 12 months (mean 1.6 ± 1.5 injuries), with injuries occurring predominantly during training and typically via non-contact mechanisms (27). In Spanish championship aerobic gymnasts (junior and senior), 90.47% reported at least one injury, most often of moderate severity, with a predominance of lower-limb involvement (65.78%) and common diagnoses including sprain and muscle strain/tear patterns (28). In contrast, among younger aerobic gymnasts (9–17 years), 25% sustained injuries during a season (10 injuries in 40 athletes), suggesting that injury occurrence may scale with exposure and/or competitive demands across development and performance levels (29). Studies in broader gymnastics samples also emphasized high prevalence of musculoskeletal problems, particularly lower-limb pain/injury, reinforcing the likelihood that repetitive landing and impact exposures remain a dominant burden pathway across gymnastic sport participation (30–32).

The anatomical distribution and injury types reported across studies were coherent with the mechanical demands of aerobic gymnastics routines and training. Lower-limb involvement was prominent across elite aerobic gymnastics samples (27, 28) and in mixed gymnastics cohorts, where ankle and knee problems were particularly frequent (30, 31). In a multi-discipline sample, ankle injuries comprised 25.5% of injuries, with knee injuries and low-back injuries also prominent; pain prevalence was high (74.4%), with low back pain reported by 35.8% of gymnasts (32). These distributions are clinically relevant because ankle and knee injuries in jumping/landing sports often reflect modifiable loading patterns and neuromuscular control demands, while lumbar complaints may reflect cumulative spinal loading, technique, and fatigue management demands during repetitive extension/rotation tasks. Importantly, several studies also noted that injuries were more frequent in training than competition (27, 28), focusing attention on daily training content, repetition volume, and fatigue accumulation as core prevention targets.

Findings regarding when and how injuries occur further support a prevention focus on high-impact technical elements and the training environment. In youth aerobic gymnasts, all injuries were reported to occur during specific technical training, with approximately half occurring after jump elements (29). In Spanish championship aerobic gymnasts, injury indices were reported as higher during training than competition, and injuries were linked particularly to difficulty elements (28). In related acrobatic gymnastics data, most recent injuries occurred during training (75.7%), with many occurring during group skills (58.8%) (33). Although acrobatic gymnastics is not aerobic gymnastics, the convergence of “training-dominant” injury occurrence and skill-execution contexts across gymnastic sports suggests that prevention strategies should prioritize risk-managed progression of technical skill load, structured landing/jump mechanics coaching, and targeted load management during high-repetition technical blocks—particularly when athletes are exposed to fatigue and when training environments introduce additional constraints (e.g., surface characteristics, equipment).

Training exposure and workload-related factors were among the clearest domains linked to injury outcomes, consistent with the review’s second objective. In youth aerobic gymnasts, injury counts were significantly associated with experience, training days, and number of competitions, and injured versus uninjured athletes differed by competition number (34). In a broader gymnastics sample, injured gymnasts trained more hours per week than uninjured gymnasts (19.9 ± 13.4 vs. 15.4 ± 11.5; p = 0.026; d = 0.36), suggesting that absolute exposure may differentiate injury risk even when the directionality and confounding structure cannot be fully resolved (32). Critically, exposure (dose) and physical capacity/readiness likely interact, such that higher training volumes may increase injury risk primarily when the imposed load exceeds the athlete’s current fitness, tissue tolerance, and recovery resources (4). Contemporary workload–injury frameworks explicitly model this dual pathway, where workloads can increase risk via fatigue while simultaneously building protective fitness adaptations, implying that the same external dose may be harmful or protective depending on the athlete’s capacity and the rate of progression (4). Therefore, an apparent training volume and following injury association in retrospective or cross-sectional designs may be partially mediated or moderated by fitness (e.g., stronger/more fatigue-resistant athletes tolerate higher chronic loads with lower relative risk), while less-prepared athletes accrue disproportionate fatigue and tissue stress at the same volume (35). This supports prevention models that pair exposure monitoring with longitudinal capacity profiling (strength-endurance, power, landing control, and fatigue resistance), rather than attempting to define a single safe volume threshold across athletes.

Physical fitness and anthropometric/alignment correlates were most clearly represented by findings relating lower-limb alignment and load distribution to injury status in aerobic gymnasts. In two related aerobic-gymnastics cohorts, injured versus uninjured gymnasts differed on right and left Q-angle measures and on bilateral weight-bearing measures, suggesting that frontal-plane alignment and asymmetrical loading may be associated with injury occurrence (34, 36). For instance, Abalo-Núñez et al. (36) reported that Q-angle (particularly the left Q-angle) and right-leg bilateral weight-bearing contributed to injury classification performance, and that the association between left Q-angle and injury varied as a function of body weight—an interaction consistent with the concept that alignment-related tissue loading is magnified under higher absolute loads. However, despite these signals, the proposed models demonstrated limited sensitivity (e.g., 58.3 and 41.7%), implying substantial false-negative rates—i.e., many athletes who later sustain injury would not be flagged—so these tools should not be used for clearance, return-to-sport permissioning, or exclusion decisions (34, 36). This limitation is consistent with broader sports-injury screening critiques showing that even statistically significant risk factors often yield poor individual-level prediction because risk distributions overlap heavily between those who will and will not be injured (37).

Psychological and pain-related correlates emerged as an important complementary domain, indicating that injury risk and burden in aerobic gymnastics may be best interpreted using a biopsychosocial framework rather than a purely biomechanical one. In world championship aerobic gymnasts, psychological stress was associated with injury presence (p = 0.043) (27), and in a broader gymnastics cohort, injured gymnasts reported higher pain catastrophizing scores than uninjured gymnasts (PCS 24.9 ± 9.4 vs. 17.8 ± 9.4; p < 0.001), while catastrophizing correlated with peak pain intensity and pain distribution (32). Although the latter sample was not isolated to aerobic gymnastics, the convergence of stress-related and pain-coping constructs with injury/pain outcomes underscores a plausible pathway whereby psychological load interacts with physical load, influencing fatigue, recovery, symptom amplification, and reporting behavior.

The prevention and health-service utilization findings also have direct practical implications. In adolescent elite gymnasts sampled across gymnastic sports, immediate professional care after problem onset was sought by fewer than 29%, primary physiotherapy prevention was reported by only 4%, and secondary prevention uptake was under 18% (30, 31). In aerobic championship gymnasts, the proportion injured was higher among those not using protective materials, and a large proportion reported requiring physiotherapy/rehabilitation services (28). Similarly, in world-level aerobic gymnasts, variables related to individual protective equipment were associated with injury status (27). Data suggest that prevention capacity may be limited not only by training exposures but also by healthcare access pathways, education, and practical prevention resources (e.g., protective measures and surface considerations).

A further domain, predictive analytics and injury “early warning,” is developing rapidly but, within the included evidence base, remains insufficiently mature for confident applied use. Traditional regression-based models in aerobic gymnasts achieved high specificity but consistently low sensitivity, suggesting limited utility for identifying the majority of athletes who will go on to be injured (34, 36). Machine-learning studies reported comparatively higher classification performance metrics, including an infrared thermal imaging approach with accuracy values reported in the 80–87% range depending on the model (38), and computer-vision/big-data approaches reporting moderate-to-high percentage performance in classifying strain/sprain categories (39). Another approach emphasized Functional Movement Screen (FMS) integration and reported improved performance when FMS inputs were included, though results were presented primarily via qualitative curve comparisons rather than comprehensive validation measures (40). Despite their promise, these algorithmic studies typically lacked the transparency required for prognostic interpretation (e.g., clear injury definitions and ascertainment, dataset provenance, robust external validation, and confounder handling). Consequently, their findings should be viewed as hypothesis-generating and methodological demonstrations rather than field-ready injury-risk tools for aerobic gymnastics at present.

Across the included evidence base, the dominant methodological limitation was the frequent use of retrospective, self-reported ascertainment for both injury outcomes and candidate risk factors, which increases vulnerability to recall bias, exposure misclassification, and outcome misclassification. This was particularly evident in studies that collected injuries at the end of a season or across long recall windows using questionnaires/interviews (27–33, 36). Relatedly, several studies used convenience, event-based, or volunteer sampling, which may have introduced selection bias and limited external validity (28–32, 36). Moreover, there was meaningful heterogeneity in injury definitions and outcome operationalisation, ranging from time-loss/disruption thresholds to current or persisting problems and reconstructed historical pain/injury experiences, which likely reduced comparability and contributed to variability in observed associations (30–33, 36). A further limitation is that our database coverage did not include SPORTDiscus or CINAHL and because gymnastics and sports-medicine content can be indexed in those sources, some relevant discipline-specific studies may not have been captured despite reference-list screening.

A second cross-cutting limitation was the insufficient handling of confounding and limitations in analytic robustness, which restricts causal/prognostic inference and complicates translation into prevention practice. Multiple studies were largely descriptive or relied on bivariate association testing without clearly prespecifying and adjusting for key confounders such as prior injury, exposure time/training load, competitive level, and age/maturation (27–32). Where regression-based prediction was attempted, concerns remain regarding small sample sizes, retrospective outcome ascertainment, and model-building practices that increase the risk of overfitting and unstable estimates (33, 34, 36). In addition, several included studies were primarily algorithm-development studies in which the injury ground truth, dataset provenance, and validation strategy were insufficiently transparent for prognostic interpretation within an epidemiologic framework, limiting confidence in clinical or field deployment (38–40). These limitations stress the need for prospective injury surveillance with standardized outcome definitions, objective exposure quantification, prespecified confounder sets, and appropriately validated models to strengthen the certainty and usability of findings in aerobic gymnastics contexts. Future research should prioritize adequately powered prospective cohort surveillance in aerobic gymnastics using standardized injury definitions and exposure denominators, repeated in-season measurement of candidate risk factors, and transparent multivariable modeling with internal and external validation; promising algorithmic approaches (thermal imaging, computer vision, and deep learning) require rigorous dataset provenance, clear injury ascertainment, and real-world validation before applied adoption.

Minimum criteria for interpreting or translating prediction/early-warning models in aerobic gymnastics should include an explicit injury/label definition with ascertainment method and time window, prospective or near-real-time outcome capture where feasible, clear separation of training and test data with internal validation and reporting of calibration, and external validation in an independent sample prior to field deployment. For machine-learning models specifically, reporting should follow contemporary extensions for prediction models to ensure transparency about data provenance, model specification, and performance evaluation.

Future research should shift from single-factor association testing toward multifactorial, time-updated risk modeling that explicitly represents how exposure, fatigue, and fitness/readiness evolve across the season. This roadmap implies standardized prospective injury surveillance with clear case definitions and short recall windows, repeated in-season capture of training dose (volume, density, and change metrics) alongside capacity markers (strength-endurance, power, landing control) and recovery/context (sleep, stress, access to care), and multivariable models that report calibration and undergo internal/external validation before applied use.

Practical implications should be framed cautiously given the predominance of retrospective designs and heterogeneous injury definitions across the evidence base. Nevertheless, the available findings support prioritizing multifactorial prevention strategies that combine structured exposure monitoring (training hours, competition density, and abrupt load changes) with technique- and capacity-focused interventions targeting high-impact elements (jump/landing mechanics and fatigue-aware session design), because injuries were reported predominantly in training and commonly linked to technical/difficulty elements and higher exposure indices (27–29, 32, 33). Screening for lower-limb alignment/asymmetry (e.g., Q-angle and weight-bearing distribution) may contribute to individualized risk profiling, but should not be used in isolation given limited sensitivity of proposed models (34, 36). The consistent signals for psychosocial correlates (stress and catastrophizing) and low prevention/early-care uptake in youth samples also justify integrating psychologically informed education, early reporting pathways, and routine access to physiotherapy/medical oversight (27, 30–32).