The Silent Saboteurs of Athletic Performance: Explainable AI Highlights Spinal Alignment and COVID-19 as Key Determinants

The Silent Saboteurs of Athletic Performance: Explainable AI Highlights Spinal Alignment and COVID-19 as Key Determinants | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Silent Saboteurs of Athletic Performance: Explainable AI Highlights Spinal Alignment and COVID-19 as Key Determinants Iman Akef Khowailed, Ibrahim M. Moustafa, Dalia Ibrahim Hemdan, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8663548/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Athletic performance arises from the interaction of biomechanical alignment, demographic characteristics, and systemic health status; however, the relative contribution of these factors remains incompletely understood. This study investigated whether sagittal cervical alignment, demographic variables, and COVID-19 exposure history jointly predict athletic performance outcomes using an explainable machine-learning framework. A retrospective analysis was conducted on pooled individual-level data from three previously published studies, comprising 562 collegiate athletes. Six predictors age, body mass index (BMI), sex, sport category, craniovertebral angle (CVA), and COVID infection count were used to model four standardized performance outcomes: agility (T-test), vertical jump height, static balance (Stork test), and dynamic balance (Y-Balance). Linear regression, random forest, and gradient boosting models were trained and evaluated, with explainability assessed using SHapley Additive exPlanations (SHAP). Across all outcomes, linear regression consistently outperformed tree-based models, indicating predominantly linear relationships between predictors and performance measures. Vertical jump and Y-Balance outcomes demonstrated strong predictability (R² ≥ 0.86), whereas agility showed weaker model performance (R² ≤ 0.35). Explainability analysis identified CVA as a dominant predictor of balance and power outcomes, while BMI and age contributed meaningfully to performance variability. COVID infection history exerted smaller but detectable effects, particularly for agility. Subgroup analyses revealed that COVID-exposed athletes and those with forward head posture relied more on postural and anthropometric predictors, whereas athletes with normal head posture demonstrated more stable predictor–outcome relationships. These findings indicate that sagittal cervical alignment is a robust neuromechanical contributor to balance and power-related athletic performance, while agility depends more strongly on unmeasured neuromotor and perceptual factors. Integrating posture-based metrics within explainable machine-learning models provides interpretable insights into performance variability and highlights the importance of considering both structural alignment and systemic health status in athlete assessment. Health sciences/Health care Health sciences/Medical research Biological sciences/Neuroscience Athletic Performance Postural Balance Machine Learning COVID-19 Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Athletic performance emerges from the interaction of biological, neuromechanical, demographic, and environmental factors that collectively influence agility, balance, and lower limb power. Conventional predictors such as age, sex, and body mass index (BMI) provide valuable population-level insights; however, these variables account for only a limited proportion of performance variability, particularly within elite or homogeneous athletic cohorts [ 1 ]. This limitation reflects the fundamentally nonlinear nature of the human motor system, in which neuromuscular drive, postural control, and task execution arise from high-dimensional interactions between central and peripheral mechanisms. Consequently, linear models based primarily on demographic descriptors are insufficient to represent the complex structure underlying athletic performance. Sagittal postural alignment has recently emerged as a critical neuromechanical determinant of motor performance. Forward head posture (FHP), commonly quantified using the craniovertebral angle (CVA), is not merely a structural deviation associated with cervical discomfort. Rather, it reflects altered integration of proprioceptive, vestibular, and visual inputs essential for sensorimotor control [ 2 ]. Clinical evidence demonstrates that correction of sagittal cervical alignment restores proprioceptive feedback, improves nerve root conduction, normalizes autonomic function, and modulates central pain processing [ 3 , 4 ].Importantly, these effects persist in long-term follow-up studies, supporting the concept that posture functions as a regulator of central nervous system organization rather than a static musculoskeletal parameter. Cervical sagittal alignment varies systematically with demographic factors and lifestyle behaviors, each exerting distinct biomechanical and sensorimotor consequences. Cross-sectional studies show that increasing age and higher BMI are associated with reduced CVA (greater forward head posture), diminished neck muscle endurance, and impaired proprioceptive acuity [ 5 ]. These demographic effects are further shaped by occupational and behavioral exposures. Prolonged sitting, excessive screen and smartphone use, and inadequate ergonomic environments have been consistently linked to increased forward head posture and neck pain, particularly among office and information technology workers [ 6 ]. During the COVID-19 pandemic, widespread shifts toward sedentary behavior, remote work, and prolonged screen exposure were associated with a population-level increase in postural deviations and neck-related complaints [ 7 ]. Beyond clinical and occupational settings, sagittal cervical alignment has demonstrated relevance in athletic performance. Our group has shown that cervical sagittal alignment predicts anticipatory core muscle activation, a key prerequisite for explosive movement and rapid change-of-direction tasks [ 8 ]. In controlled athletic cohorts, forward head posture has been associated with reduced agility, lower limb power deficits, shorter single-leg stance duration, and decreased Y-balance composite scores. These functional impairments were accompanied by altered somatosensory evoked potentials, indicating degraded sensorimotor integration [ 9 ]. Collectively, these findings support the conceptualization of posture as a neuromechanical input variable that shapes the perceptual reference frame from which movement strategies are generated. In parallel, systemic health factors introduce additional variability into athletic performance capacity. SARS-CoV-2 exposure represents a biologically plausible but underexplored modifier of neuromuscular performance. Empirical evidence indicates that COVID-19 infection can exert persistent negative effects on balance, movement kinetics, neuromuscular function, and central fatigue, even in young and previously healthy athletic populations [ 10 , 11 ]. Given the high prevalence of COVID-19 exposure in contemporary athletes, infection burden may meaningfully influence performance outcomes. Several post–SARS-CoV-2 sequelae relevant to athletic performance have been identified. Alterations in autonomic regulation, including reduced or dysregulated heart rate variability (HRV), suggest impaired autonomic control that may compromise exercise tolerance and recovery [ 12 ]. Mechanistic studies link mitochondrial dysfunction to long-COVID, providing a biological basis for persistent fatigue and reduced oxidative capacity. Objective assessments demonstrate balance deficits following mild infection, altered running mechanics characterized by prolonged stance time and reduced propulsion, and neuromuscular impairments including prolonged fatigue and reduced power output [ 13 ]. These findings imply that previously infected athletes may exhibit reduced tolerance to cumulative mechanical stress and require individualized, closely monitored return-to-training strategies. Despite the growing recognition of posture, demographic factors, and systemic health as determinants of athletic performance, existing predictive models overwhelmingly rely on linear regression or single-domain indicators. Such approaches fail to capture nonlinear interactions between neuromechanical alignment, demographic characteristics, and biological stressors. Machine learning (ML) and artificial intelligence (AI) methods offer a powerful alternative by modeling complex, multidimensional dependencies and have demonstrated superior predictive performance in sport analytics, injury prediction, and talent identification[ 14 , 15 ]. However, current AI applications in sports science rarely integrate postural biomarkers or COVID-related variables, and few provide interpretable outputs capable of informing clinical or training decisions. Explainable Artificial Intelligence (XAI) addresses this limitation by enabling transparency and interpretability in ML models. XAI techniques identify influential features, clarify decision pathways, and enhance trust in model predictions [ 16 ]. Feature attribution methods are central to XAI, quantifying the contribution of each input variable to model outputs at both global and individual levels. Among these, SHapley Additive exPlanations (SHAP) applies cooperative game theory to assign contribution values to features based on their marginal impact relative to a baseline prediction [ 17 ]. XAI has increasingly been adopted in sports science to improve precision and clinical relevance in athlete assessment. Applications using SHAP have identified key physiological, biomechanical, and demographic contributors to performance and injury risk [ 18 ]. In swimming, XAI analyses revealed that body composition, training load, and lower limb power were dominant performance determinants [ 19 ]. In soccer, XAI applied to neuromuscular and biomechanical data identified predictors of muscle injury, supporting individualized prevention strategies [ 20 ]. Collectively, these findings demonstrate that XAI bridges the gap between predictive accuracy and practical applicability [ 21 ]. Accordingly, the present study hypothesizes that integrating sagittal cervical alignment, demographic characteristics, and COVID-19 exposure variables within an explainable machine learning framework will improve the prediction of athletic performance outcomes compared with conventional linear models. Furthermore, it is hypothesized that XAI-based feature attribution will identify posture-related and systemic health variables as key contributors to performance variability, providing mechanistically interpretable insights relevant to training optimization and injury risk management. Methods Study design and data sources This study employed a retrospective machine learning (ML) design using pooled, individual-level data derived from three previously published investigations conducted in collegiate athletes. The contributing datasets examined posture, demographic characteristics, and skill-related athletic performance using standardized field-based protocols. The first dataset was a 3D posture–performance study , which quantified translational and rotational displacement of spinal segments using PostureScreen® Mobile and assessed athletic performance through agility, lower-limb power, static balance, dynamic balance, and cardiopulmonary exercise testing. The second dataset comprised a forward head posture (FHP)cohort, which compared athletes with craniovertebral angle (CVA)–defined sagittal cervical misalignment to matched controls with normal alignment and reported outcomes including T-test agility, non-countermovement vertical jump, Stork static balance, and Y-balance performance. The third dataset was a Long COVID athlete cohort, which stratified athletes based on documented SARS-CoV-2 infection history and included standardized assessments of agility, vertical jump, static balance, and Y-balance performance. Only variables that were conceptually equivalent and consistently available across all three datasets were considered for pooled analysis. Ethical approval All studies were approved by the relevant institutional ethics committees in accordance with the Declaration of Helsinki, with written informed consent obtained from all participants (University of Sharjah IRB: REC-22-11-26-S; University of Sharjah IRB: REC-22-06-09-03; Cairo University Research Ethics Committee, Faculty of Physical Therapy: P.T.Rec/012/0050051). The present work involved secondary analysis of fully de-identified data, with no participant contact or new data collection. Data harmonization and analytic sample To ensure methodological consistency and avoid selective availability bias, a c ommon-variable framework was applied. Variables not uniformly available across all datasets were excluded a priori. Accordingly, neurophysiological measures (e.g., somatosensory evoked potential components), radiographic alignment parameters, and cardiopulmonary variables were not included in the current analysis. After harmonization, quality screening, and removal of incomplete records, the final analytic sample comprised 562 collegiate athletes . Predictor variables Six predictor variables were incorporated into the machine learning models: age (years), body mass index (BMI; kg/m²), COVID-19 count (number of documented prior SARS-CoV-2 infections), craniovertebral angle (CVA; degrees), sex (binary encoded), and sport category (binary encoded). CVA was treated as a continuous biomechanical variable and also used to derive a posture classification. Forward head posture was defined as CVA < 50° , consistent with established reliability thresholds, whereas CVA ≥ 50° indicated non-forward head posture. COVID exposure was determined from documented infection history reported in the original datasets, and COVID count reflected cumulative infection burden where applicable. Outcome variables Four standardized field-based indicators of athletic performance were modeled as continuous outcome variables: agility (T-test time, seconds), vertical jump height (cm), static balance (Stork test, seconds), and dynamic balance (Y-Balance composite score, %). These outcomes were selected to represent complementary domains of athletic performance, including neuromuscular coordination, explosive power, postural control, and dynamic balance. Data preprocessing All datasets were merged into a single analytic table following variable renaming, unit verification, and range screening. Continuous predictors were examined for implausible values and distributional anomalies. For ML modeling, numeric predictors were scaled using statistics derived exclusively from the training data , with identical transformations applied to the test set to prevent data leakage. Categorical predictors were encoded using predefined binary schemes applied consistently across cohorts. Missing data handling was prespecified. When sporadic missing values were present, imputation was performed using training-set–derived statistics only. Machine learning model development Given the anticipated nonlinear and interactive relationships among determinants of athletic performance, a comparative machine learning framework was implemented. The models evaluated included regularized linear regression as a baseline reference model, random forest regression, and gradient boosting regression. The dataset was randomly partitioned into a training set (80%) and an independent test set (20%) . Hyperparameter tuning was conducted within the training set using k-fold cross-validation . Model selection was based on cross-validated performance, and final model evaluation was performed on the held-out test set. Model performance was quantified using the coefficient of determination ( R² ) as the primary metric, supplemented by absolute error measures (e.g., RMSE and MAE) to support interpretability of prediction error in native units. Explainable artificial intelligence analysis To enhance interpretability, SHapley Additive exPlanations (SHAP) were applied to the final models. Feature attribution was examined at both the global level (mean absolute SHAP values) and the individual level to identify the relative contribution of posture, demographic, and COVID-related variables to each performance outcome. Dependence plots were used to explore linear and nonlinear relationships between predictors and outcomes. Subgroup and sensitivity analyses Prespecified subgroup analyses were conducted to examine whether posture–performance relationships differed by COVID exposure status and posture classification (FHP vs non-FHP). Sensitivity analyses were performed to confirm the stability of predictor importance rankings across subgroups. Software and reproducibility All analyses were performed using Python-based machine learning libraries. A fixed random seed was applied to data splitting, model training, and cross-validation procedures to ensure reproducibility. The full preprocessing and modeling pipeline was implemented as a unified workflow to minimize analytic variability across outcomes. Data Availability All data generated and/or analyzed during the current study are included in the published article and its supplementary information files. Results 3.1 Data Preparation and Descriptive Results Following data cleaning and harmonization procedures, a total of 562 participants were retained for the final analysis. Records containing missing values, implausible entries, or empty variables were excluded to ensure data integrity. All numerical predictors were standardized prior to model training to place variables on comparable scales and to facilitate stable and unbiased machine-learning estimation. Descriptive statistics for demographic, postural, and athletic performance variables are summarized in Table 1 . The cohort represented a relatively homogeneous athletic population, characterized by a narrow age range and body mass index distribution. This homogeneity reduces demographic confounding and allows model performance to more directly reflect neuromechanical and postural influences rather than broad population variability. Craniovertebral angle (CVA) values spanned both forward head posture and normal alignment ranges, providing sufficient variability to support posture-based subgroup analyses and interpretable modeling of posture–performance relationships. Table 1. Descriptive statistics for demographic, postural, and performance variables. Variable Mean ± SD Min–Max n (%) Age (years) 22.46 ± 1.94 17–26 562 BMI (kg/m²) 22.15 ± 1.51 18.50–24.90 562 COVID_Count 1.04 ± 0.56 0–3 562 CVA (°) 49.76 ± 2.99 41.96–57.54 562 Agility_TTest_s 9.70 ± 1.11 6.00–13.00 562 Vertical_Jump_cm 40.44 ± 1.77 35.21–44.37 562 Stork_Balance_s 23.97 ± 1.38 20.90–27.98 562 Y_Balance_% 85.12 ± 1.56 81.07–89.76 562 Sex – Female — — 283 (50.4%) Sex – Male — — 279 (49.6%) Sport Category – Lower Limb — — 426 (75.8%) Sport Category – Upper Limb — — 136 (24.2%) Subgroup stratification was performed based on COVID exposure status and postural alignment to enable targeted comparative modeling. Sixty-one participants were classified as COVID-free (COVID_Count = 0), whereas 501 participants had a documented history of one or more SARS-CoV-2 infections. Based on established craniovertebral angle thresholds, 291 participants were categorized as exhibiting forward head posture (CVA < 50°), and 271 were classified as having normal head posture (CVA ≥ 50°). Subgroup distributions are presented in Table 2 . These stratifications allowed examination of whether posture–performance relationships and predictor importance patterns differed according to systemic health status and sagittal alignment, thereby supporting interpretable subgroup-specific machine-learning analyses. Table 2. Sample distribution for each subgroup. Subgroup Definition n COVID-Free COVID_Count = 0 61 COVID-Exposed COVID_Count ≥ 1 501 Forward Head CVA < 50° 291 Normal Head CVA ≥ 50° 271 Total - 562 3.2 Model Performance Across Outcomes Three machine-learning models— Linear Regression, Random Forest, and XGBoost —were trained for each athletic performance outcome to compare linear and nonlinear modeling approaches. Model performance metrics, including the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE), are summarized in Table 3 . Across all outcomes, Linear Regression consistently achieved the highest explanatory power and lowest prediction error , outperforming tree-based models. This consistent pattern indicates that the dominant relationships linking postural alignment, demographic variables, and athletic performance in this cohort were predominantly linear in nature, with limited evidence of strong nonlinear interactions. These findings suggest that, within a relatively homogeneous athletic population, performance variability is largely governed by proportional biomechanical and physiological relationships rather than complex hierarchical feature interactions. Table 3. Model performance (R², RMSE, MAE) for all outcomes. Outcome Model R² RMSE MAE Agility Linear Regression 0.346 0.834 0.628 Agility Random Forest 0.193 0.927 0.686 Agility XGBoost 0.089 0.984 0.724 Vertical Jump Linear Regression 0.893 0.573 0.351 Vertical Jump Random Forest 0.872 0.627 0.422 Vertical Jump XGBoost 0.859 0.656 0.439 Stork Balance Linear Regression 0.767 0.704 0.471 Stork Balance Random Forest 0.714 0.78 0.528 Stork Balance XGBoost 0.711 0.784 0.567 Y-Balance Linear Regression 0.869 0.57 0.339 Y-Balance Random Forest 0.843 0.624 0.394 Y-Balance XGBoost 0.83 0.649 0.435 3.2.1 Agility (T-Test; lower is better) Agility performance demonstrated the weakest predictability across all models. Linear Regression achieved an R² of 0.35 with an RMSE of 0.83, whereas Random Forest and XGBoost showed lower explanatory power (R² = 0.19 and 0.08, respectively) and higher prediction error. SHAP-based explainability results for agility are presented in Figure 1a–b . Predictor contributions were relatively diffuse, with no single anthropometric or postural variable exerting dominant influence. This pattern suggests that agility performance is only partially explained by static structural characteristics and likely reflects a greater dependence on higher-order neuromotor, perceptual, and decision-making processes that are not captured by the current input features. Figure 1. SHAP summary (a) and dependence (b) plots for agility (T-test) performance across all participants. Feature attribution demonstrates low and diffuse predictor influence, with no single demographic or postural variable exerting dominant control. This pattern is consistent with the modest predictive performance observed for agility outcomes and reflects the complex, open-skill nature of agility tasks. 3.2.2 Vertical Jump (higher is better) Vertical jump height exhibited the strongest predictability among all outcomes. Linear Regression achieved the highest coefficient of determination (R² = 0.89) with the lowest RMSE (0.57), followed by Random Forest (R² = 0.87, RMSE = 0.62) and XGBoost (R² = 0.86, RMSE = 0.65). SHAP plots for this outcome are shown in Figure 2a–b , demonstrating concentrated feature attribution. The high predictive accuracy and focused SHAP contributions indicate that vertical jump performance is strongly governed by proportional biomechanical relationships linking sagittal alignment and body composition to lower-limb power generation. Figure 2. SHAP summary (a) and dependence (b) plots for vertical jump height across all participants. Craniovertebral angle (CVA) and body mass index (BMI) emerged as the dominant contributors, indicating that sagittal cervical alignment and body composition strongly influence explosive lower-limb power generation. 3.2.3 Stork Balance For static balance performance, Linear Regression achieved an R² of 0.77 with an RMSE of 0.70, outperforming Random Forest and XGBoost (both R² ≈ 0.71, RMSE ≈ 0.78). SHAP plots for Stork balance are presented in Figure 3a–b . Balance performance showed strong linear relationships, particularly with age and craniovertebral angle. The monotonic nature of these relationships indicates that static postural stability is primarily governed by proportional neuromechanical control mechanisms, with limited contribution from higher-order nonlinear interactions. Figure 3. SHAP summary (a) and dependence (b) plots for static balance (Stork test). Age and craniovertebral angle demonstrated consistent, monotonic relationships with balance duration, reflecting predominantly linear neuromechanical control mechanisms underlying static postural stability. 3.2.4 Y-Balance Y-Balance performance was strongly predictable across all models. Linear Regression achieved the highest R² (0.87) with the lowest RMSE (0.56), compared with Random Forest (R² = 0.84, RMSE = 0.62) and XGBoost (R² = 0.83, RMSE = 0.64). SHAP plots for this outcome are shown in Figure 4a–b , indicating substantial contributions from anthropometric and postural indicators. The concentrated feature attribution and strong linear model performance suggest that dynamic balance during multi-directional reach tasks is closely linked to sagittal alignment and body composition, reflecting structured neuromechanical control rather than highly adaptive or nonlinear motor strategies. Figure 4. SHAP summary (a) and dependence (b) plots for Y-Balance performance. Craniovertebral angle exhibited the strongest contribution, highlighting the role of sagittal alignment in dynamic postural control and reach stability during multi-directional balance tasks. 3.3 Subgroup Modeling Separate models were developed for each predefined subgroup to evaluate whether predictor–outcome relationships differed according to COVID exposure status and postural alignment. Performance comparisons across subgroups are summarized in Table 4 . Across all subgroups and outcomes, Linear Regression consistently remained the best-performing model , indicating that subgroup stratification primarily influenced model stability and explanatory strength rather than altering the underlying structure of predictor–outcome relationships. Table 4. Performance comparisons across subgroups. Subgroup Outcome Best R² Best Model COVID-Free Agility 0.21 Linear Regression COVID-Free Vertical Jump 0.81 Linear Regression COVID-Free Stork Balance 0.62 Linear Regression COVID-Free Y-Balance 0.79 Linear Regression COVID-Exposed Agility 0.33 Linear Regression COVID-Exposed Vertical Jump 0.89 Linear Regression COVID-Exposed Stork Balance 0.75 Linear Regression COVID-Exposed Y-Balance 0.87 Linear Regression Forward Head Agility 0.22 Linear Regression Forward Head Vertical Jump 0.86 Linear Regression Forward Head Stork Balance 0.71 Linear Regression Forward Head Y-Balance 0.83 Linear Regression Normal Head Agility 0.36 Linear Regression Normal Head Vertical Jump 0.9 Linear Regression Normal Head Stork Balance 0.78 Linear Regression Normal Head Y-Balance 0.88 Linear Regression Table 4. Performance comparisons across subgroups. 3.3.1 COVID-Free vs COVID-Exposed COVID-free participants demonstrated weaker model fits across all outcomes, reflected by lower R² values. This pattern likely reflects increased internal variability and reduced signal-to-noise ratio within the smaller COVID-free subgroup (n = 61), rather than a fundamental change in predictor relevance. In contrast, COVID-exposed participants exhibited more stable and consistent model performance across outcomes, suggesting that athletic performance in this group was more tightly constrained by measurable postural and anthropometric factors. SHAP-based subgroup explainability results are presented in Figure 5a–d . Figure 5. SHAP summary plots stratified by COVID exposure status for all performance outcomes. In COVID-exposed athletes, postural and anthropometric predictors showed amplified influence compared with COVID-free participants, suggesting reduced physiological buffering capacity and greater sensitivity to biomechanical inefficiencies following infection. 3.3.2 Forward Head vs Normal Head Posture Participants with forward head posture demonstrated lower predictability for balance and agility outcomes, indicating greater internal variability and disrupted neuromechanical organization within this subgroup. In contrast, the normal head posture subgroup exhibited higher R² values and more consistent biomechanical patterns across outcomes. These findings suggest that preserved sagittal alignment is associated with more stable and coherent posture–performance relationships. Posture-based SHAP comparisons are visualized in Figure 6a–d . Figure 6. SHAP summary plots stratified by head posture classification. Athletes with forward head posture demonstrated greater variability and reduced predictability across balance and agility outcomes, whereas those with normal head posture exhibited more stable and coherent predictor–outcome relationships, indicating more efficient neuromechanical organization. 3.4 Explainability Analysis (SHAP Results) 3.4.1 Global Findings (Figures 1–4) Across all models and athletic performance outcomes, SHAP analysis revealed a highly consistent hierarchy of predictor importance , indicating robust and stable feature–outcome relationships. Craniovertebral angle (CVA) consistently emerged as one of the most influential predictors, particularly for balance-related outcomes and vertical jump performance, underscoring the central role of sagittal cervical alignment in neuromechanical control and force transmission. Body mass index (BMI) demonstrated a stable and meaningful contribution to agility and vertical jump outcomes, reflecting the influence of body composition on movement efficiency and power generation. Age showed a clear negative association with balance performance, consistent with age-related changes in postural control mechanisms. COVID_Count exhibited measurable but comparatively smaller effects overall, with its strongest influence observed for agility performance, suggesting sensitivity of open-skill tasks to systemic health perturbations. Global SHAP summary and bar plots illustrating these relationships are presented in Figures 1–4 . 3.4.2 Subgroup SHAP Patterns (Figures 5–6) Subgroup-specific SHAP analyses revealed systematic shifts in predictor dominance , indicating that systemic health status and postural alignment modulate the relative contribution of performance determinants rather than introducing new predictors. In COVID-free participants, age emerged as the dominant contributor across outcomes, suggesting that in the absence of infection-related stressors, performance variability is primarily governed by baseline physiological maturation and neuromuscular efficiency. In contrast, among COVID-exposed participants, CVA and BMI gained greater relative importance, indicating increased reliance on structural alignment and anthropometric factors when physiological buffering capacity may be reduced. Posture-based subgroup analyses further demonstrated that within the forward head posture subgroup, CVA exerted the strongest negative contribution to balance outcomes, reflecting disrupted neuromechanical organization and reduced postural stability. Conversely, in the normal head posture subgroup, predictor influences were more evenly distributed, consistent with a more integrated and resilient neuromotor control system. Subgroup SHAP visualizations are shown in Figures 5 and 6 . 3.5 Summary of Key Findings Vertical jump and Y-Balance outcomes demonstrated strong model performance (R² ≥ 0.86), indicating that these performance domains are highly predictable from postural and anthropometric features. In contrast, agility exhibited weaker predictability, suggesting that additional neuromotor, perceptual, or cognitive variables may be required to fully capture the determinants of open-skill athletic tasks. Explainability analysis consistently identified craniovertebral angle, body mass index, and age as the most influential predictors across outcomes. Subgroup analyses further demonstrated that COVID history and head posture modulate the strength and stability of these relationships , highlighting the importance of integrating both systemic health status and sagittal alignment when interpreting athletic performance. Discussion This study employed interpretable machine-learning models to investigate how sagittal cervical alignment, demographic factors, and COVID-19 exposure history jointly relate to athletic performance in a homogeneous cohort of collegiate athletes. By integrating individual-level data from multiple previously published investigations, the present analysis extends prior work by providing a unified predictive and explainable framework linking posture-related variables to balance, power, and agility outcomes. The principal findings indicate that craniovertebral angle (CVA), body mass index (BMI), and age are consistently associated with balance and power-related performance, whereas agility performance is less strongly explained by static anthropometric and postural features. Model Behavior and Predictive Structure Across all performance outcomes, linear regression consistently outperformed Random Forest and XGBoost models, yielding higher explained variance and lower prediction error. This finding suggests that, within this relatively homogeneous athletic cohort, the dominant relationships between posture, demographic variables, and athletic performance are primarily linear. Although neuromuscular control is inherently nonlinear, the current results indicate that such nonlinearities were either weak, masked by measurement noise, or not captured by the available feature set. Similar observations have been reported in athletic and biomechanical datasets characterized by restricted demographic variability and limited feature dimensionality, where linear models often outperform more complex approaches[22,23]. Importantly, the superior performance of linear models does not imply the absence of nonlinear physiological mechanisms. Rather, it suggests that proportional biomechanical relationships—such as those linking sagittal alignment to force transmission and postural stability—account for a substantial proportion of observable performance variability in this cohort. The absence of high-dimensional neuromotor, cognitive, or training-load variables likely constrained the capacity of tree-based models to exploit nonlinear interactions. Outcome-Specific Interpretation Vertical jump and balance-related outcomes demonstrated strong predictability, with coefficients of determination exceeding 0.77 across models. Explainability analysis revealed that CVA and BMI were dominant contributors to vertical jump performance, supporting the notion that sagittal cervical alignment and body composition influence lower-limb power generation through biomechanical coupling and postural organization [24–26]. These findings align with prior experimental and clinical studies demonstrating that altered cervical alignment is associated with impaired neuromuscular coordination and reduced force efficiency [4,27]. Static and dynamic balance outcomes (Stork balance and Y-Balance) were also strongly associated with CVA and age. The monotonic relationships observed in SHAP dependence plots suggest that balance performance is governed by proportional neuromechanical control mechanisms rather than highly adaptive or nonlinear strategies. This observation is consistent with previous work linking sagittal alignment to sensorimotor integration, proprioceptive accuracy, and postural stability[28–30]. In contrast, agility performance exhibited substantially weaker predictability across all models. SHAP attribution patterns were diffuse, with no single predictor exerting dominant influence. Agility tasks are inherently open skill in nature, requiring rapid perceptual processing, anticipatory control, and decision-making under time constraints. These higher order neuromotor and cognitive components are not captured by static anthropometric or postural measures, which likely explains the limited explanatory power observed in the present models. Similar limitations have been reported in prior performance modeling studies, where agility outcomes required task-specific or neurocognitive inputs to achieve meaningful prediction accuracy[31,32]. Subgroup-Specific Effects of COVID Exposure and Postural Alignment Subgroup analyses revealed systematic differences in model stability and predictor importance without altering the fundamental structure of predictor–outcome relationships. COVID-free athletes demonstrated weaker model fits across outcomes, likely reflecting increased internal variability and reduced statistical power associated with the smaller subgroup size. In contrast, COVID-exposed athletes exhibited more stable and consistent predictive relationships, with CVA and BMI assuming greater relative importance. These findings suggest that following systemic stressors such as SARS-CoV-2 infection, athletic performance may become more tightly constrained by measurable biomechanical and anthropometric factors, potentially reflecting reduced physiological buffering capacity[12,33,34]. Posture-based stratification further highlighted meaningful differences. Athletes with forward head posture showed lower predictability for balance and agility outcomes, indicating greater internal variability and disrupted neuromechanical organization. Conversely, athletes with normal head posture demonstrated higher explained variance and more evenly distributed predictor contributions, consistent with more integrated and resilient postural control systems. These observations align with previous reports linking forward head posture to altered sensorimotor integration and compromised postural stability[28,29,35]. Explainability and Clinical Interpretability A key strength of this study lies in the application of SHAP-based explainability methods, which enabled transparent interpretation of model predictions at both global and subgroup levels. Across all outcomes, CVA consistently emerged as one of the most influential predictors, underscoring the relevance of sagittal cervical alignment as a neuromechanical factor rather than a purely structural descriptor. BMI contributed meaningfully to power and agility outcomes, while age showed a consistent negative association with balance performance, reflecting well-established age-related changes in postural control [36]. COVID_Count demonstrated smaller but detectable effects, with its strongest influence observed in agility outcomes. This pattern suggests that open-skill tasks may be particularly sensitive to subtle systemic or neuromotor perturbations following infection, even in young athletic populations. Importantly, subgroup SHAP analyses revealed shifts in predictor dominance rather than the emergence of new predictors, indicating that systemic health status and postural alignment modulate the relative importance of performance determinants rather than redefining them. Funding: This research was funded by the Ongoing Research Funding program (ORF-2026-798), King Saud University, Riyadh, Saudi Arabia. Limitations Several limitations should be acknowledged. First, the retrospective and observational design precludes causal inference; SHAP-derived feature importance reflects association rather than mechanistic causality. Second, COVID exposure was quantified using a simplified count variable, without information on infection severity, symptom persistence, vaccination status, or time since infection. Third, pooling data from multiple studies introduces potential protocol heterogeneity despite harmonization efforts. Fourth, the absence of neuromotor, cognitive, training-load, and injury-history variables likely limited predictive performance, particularly for agility outcomes. Finally, external validation in an independent cohort is required to confirm generalizability. Implications and Future Directions Despite these limitations, the present findings demonstrate that sagittal cervical alignment and basic anthropometric characteristics are robustly associated with balance and power-related athletic performance. The results support the potential inclusion of posture-based metrics within performance screening and monitoring frameworks, while emphasizing that such measures should be interpreted as contributory rather than determinative. Future research should incorporate prospective designs, richer neuromotor feature sets, and injury outcomes to evaluate whether posture-informed, explainable machine-learning models can meaningfully inform performance optimization and injury risk stratification. Conclusion In conclusion, sagittal cervical alignment and basic anthropometric factors are strongly associated with balance and power-related athletic performance, whereas agility is less well explained by static postural and demographic variables. Explainable machine-learning analysis consistently identified craniovertebral angle as a key contributor to balance and vertical jump outcomes, with smaller but detectable effects of COVID-19 exposure history. These findings highlight the value of posture-based metrics within interpretable modeling frameworks and underscore the importance of considering both structural alignment and systemic health status when evaluating athletic performance. Declarations Competing Interests Yes, the authors have competing interests as defined by Nature Research, or other interests that could be perceived to influence the results and/or discussion reported in this paper. These will be disclosed explicitly in the Competing Interests section of the manuscript.Iman Funding: This research was funded by the Ongoing Research Funding program (ORF-2026-798), King Saud University, Riyadh, Saudi Arabia. Author Contribution I.K. and I.M. conceived and designed the study. Data curation and methodological development were performed by I.K., I.M., D.M., C.P., and E.C. Machine learning modeling and explainable AI analyses were conducted by I.M., I.K., E.C. Figures and data visualization were prepared by I.K. and I.M. Manuscript drafting and writing were carried out by I.K., D.M., C.P., I.M. and A.A. contributed to interpretation of the results and critical revision of the manuscript. All authors reviewed, edited, and approved the final version of the manuscript. Data Availability All data generated and/or analyzed during the current study are included in the published article and its supplementary information files. References Behm, D. G., Muehlbauer, T., Kibele, A. & Granacher, U. Effects of Strength Training Using Unstable Surfaces on Strength, Power and Balance Performance Across the Lifespan: A Systematic Review and Meta-analysis. Sports Med. 45 , 1645–1669 (2015). Panjabi, M. M. The Stabilizing System of the Spine. Part I. Function, Dysfunction, Adaptation, and Enhancement. J. Spinal Disord . 5 , 383–389 (1992). Moustafa, I. M., Diab, A. A., Hegazy, F. & Harrison, D. E. Does improvement towards a normal cervical sagittal configuration aid in the management of cervical myofascial pain syndrome: a 1- year randomized controlled trial. BMC Musculoskelet. Disord . 19 , 396 (2018). Moustafa, I. M., Diab, A. A. M., Hegazy, F. A. & Harrison, D. E. Does rehabilitation of cervical lordosis influence sagittal cervical spine flexion extension kinematics in cervical spondylotic radiculopathy subjects? J. Back Musculoskelet. Rehabil . 30 , 937–941 (2017). Kocur, P. et al. Relationship between age, BMI, head posture and superficial neck muscle stiffness and elasticity in adult women. Sci Rep 9 , (2019). Stincel, O. R. et al. Assessment of Forward Head Posture and Ergonomics in Young IT Professionals – Reasons to Worry? Medicina del. Lavoro 114 , (2023). Gomez, I. N., Suarez, C. G., Sosa, K. E. & Tapang, M. L. Work from home-related musculoskeletal pain during the COVID-19 pandemic: A rapid review. International Journal of Osteopathic Medicine vol. 47 Preprint at (2023). https://doi.org/10.1016/j.ijosm.2022.12.001 Moustafa, I. M., Diab, A. A. & Harrison, D. E. The Efficacy of Cervical Lordosis Rehabilitation for Nerve Root Function and Pain in Cervical Spondylotic Radiculopathy: A Randomized Trial with 2-Year Follow-Up. J. Clin. Med. 11 , 6515 (2022). Moustafa, I., Kim, M. & Harrison, D. E. Comparison of Sensorimotor Integration and Skill-Related Physical Fitness Components Between College Athletes With and Without Forward Head Posture. J. Sport Rehabil . 32 , 53–62 (2023). Jafarnezhadgero, A. A., Noroozi, R., Fakhri, E., Granacher, U. & Oliveira, A. S. The Impact of COVID-19 and Muscle Fatigue on Cardiorespiratory Fitness and Running Kinetics in Female Recreational Runners. Front Physiol 13 , (2022). Sedaghati, P., Balayi, E. & Ahmadabadi, S. Effects of COVID-19 related physical inactivity on motor skills in children with intellectual disability. BMC Public. Health 22 , (2022). Zacher, J., Branahl, A., Predel, H. G. & Laborde, S. Effects of Covid-19 on the autonomic nervous system in elite athletes assessed by heart rate variability. Sport Sci. Health . 19 , 1269–1280 (2023). Molnar, T. et al. Mitochondrial dysfunction in long COVID: mechanisms, consequences, and potential therapeutic approaches. GeroScience vol. 46 5267–5286 Preprint at (2024). https://doi.org/10.1007/s11357-024-01165-5 Taborri, J. et al. A Machine-Learning Approach to Measure the Anterior Cruciate Ligament Injury Risk in Female Basketball Players. Sensors 21 , 3141 (2021). Pietraszewski, P. et al. The Role of Artificial Intelligence in Sports Analytics: A Systematic Review and Meta-Analysis of Performance Trends. Appl. Sci. 15 , 7254 (2025). Barredo Arrieta, A. et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Inform. Fusion . 58 , 82–115 (2020). Linardatos, P., Papastefanopoulos, V. & Kotsiantis, S. Explainable AI: A Review of Machine Learning Interpretability Methods. Entropy 23 , 18 (2020). Kranzinger, S., Halmich, C., Hofer, D. & Kranzinger, C. A scoping review of explainable artificial intelligence in sports science. Discover Artif. Intell. https://doi.org/10.1007/s44163-025-00709-8 (2025). Carvalho, D. D. et al. Swimming Performance Interpreted through Explainable Artificial Intelligence (XAI)—Practical Tests and Training Variables Modelling. Applied Sci. (Switzerland) 14, (2024). Calderón-Díaz, M. et al. Explainable Machine Learning Techniques to Predict Muscle Injuries in Professional Soccer Players through Biomechanical Analysis. Sensors 24 , (2024). Souaifi, M. et al. Artificial Intelligence in Sports Biomechanics: A Scoping Review on Wearable Technology, Motion Analysis, and Injury Prevention. Bioengineering vol. 12 Preprint at (2025). https://doi.org/10.3390/bioengineering12080887 Leckey, C., Van Dyk, N., Doherty, C., Lawlor, A. & Delahunt, E. Machine learning approaches to injury risk prediction in sport: A scoping review with evidence synthesis. British Journal of Sports Medicine vol. 59 491–500 Preprint at (2025). https://doi.org/10.1136/bjsports-2024-108576 Jauhiainen, S. et al. Predicting ACL Injury Using Machine Learning on Data From an Extensive Screening Test Battery of 880 Female Elite Athletes. Am. J. Sports Med. 50 , 2917–2924 (2022). Luu, B. C. et al. Machine Learning Outperforms Logistic Regression Analysis to Predict Next-Season NHL Player Injury: An Analysis of 2322 Players From 2007 to 2017. Orthop J. Sports Med 8 , (2020). Markovic, G. & Jaric, S. Is vertical jump height a body size-independent measure of muscle power? J. Sports Sci. 25 , 1355–1363 (2007). Ben Mansour, G., Kacem, A., Ishak, M., Grélot, L. & Ftaiti, F. The effect of body composition on strength and power in male and female students. BMC Sports Sci. Med. Rehabil 13 , (2021). Tamim, M., Moustafa, I. M., Alaparthi, G. K., Oakley, P. A. & Harrison, D. E. Translational and Rotational Postural Aberrations Are Related to Pulmonary Functions and Skill-Related Physical Fitness Components in Collegiate Athletes. J Clin. Med 12 , (2023). Harrison, D. D. et al. Modeling of the Sagittal Cervical Spine as a Method to Discriminate Hypolordosis. Spine (Phila Pa. 1976) . 29 , 2485–2492 (2004). Treleaven, J. Sensorimotor disturbances in neck disorders affecting postural stability, head and eye movement control. Man. Ther. 13 , 2–11 (2008). Ha, S. Y. & Sung, Y. H. A temporary forward head posture decreases function of cervical proprioception. J. Exerc. Rehabil . 16 , 168–174 (2020). Lin, G., Zhao, X., Wang, W. & Wilkinson, T. The relationship between forward head posture, postural control and gait: A systematic review. Gait and Posture vol. 98 316–329 Preprint at (2022). https://doi.org/10.1016/j.gaitpost.2022.10.008 Sheppard, J. M. & Young, W. B. Agility literature review: Classifications, training and testing. J. Sports Sci. 24 , 919–932 (2006). Scanlan, A., Humphries, B., Tucker, P. S. & Dalbo, V. The influence of physical and cognitive factors on reactive agility performance in men basketball players. J. Sports Sci. 32 , 367–374 (2014). Guzik, A., Wolan-Nieroda, A., Kochman, M., Perenc, L. & Drużbicki, M. Impact of mild COVID-19 on balance function in young adults, a prospective observational study. Sci Rep 12 , (2022). Molnar, T. et al. Mitochondrial dysfunction in long COVID: mechanisms, consequences, and potential therapeutic approaches. GeroScience vol. 46 5267–5286 Preprint at (2024). https://doi.org/10.1007/s11357-024-01165-5 Xia, Q. et al. Factors associated with balance impairments in the community-dwelling elderly in urban China. BMC Geriatr 23 , (2023). Additional Declarations Competing interest reported. Yes, the authors have competing interests as defined by Nature Research, or other interests that could be perceived to influence the results and/or discussion reported in this paper. These will be disclosed explicitly in the Competing Interests section of the manuscript. Iman Supplementary Files SupplementaryFilesXAIData.xlsx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 15 Apr, 2026 Editor assigned by journal 31 Mar, 2026 Editor invited by journal 05 Feb, 2026 Submission checks completed at journal 04 Feb, 2026 First submitted to journal 04 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8663548","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":623636172,"identity":"04dc833f-9ddf-4cc6-aaa3-1581f5b3f4b8","order_by":0,"name":"Iman Akef Khowailed","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEElEQVRIie2PsUoDQRCGJwxcmjmv3bBy9wp7HBzaxFfZsGC1AcFGMJhUZ7NcHfA9UkcCsYn9FiJIwE4QDkJADK5ayurZCe5XzE4x3/4zAIHAHyQien/YPoCao+sQWEvF1WPZToFPBVzVAj/an5Q9ur1e0+iAkq5umpPRfTe5mnSet9BPvYvFtSpoyahnnmZ8ujxFdjfHngFVeJWESq4jRsIOZ0iRu8dK4O6mweRbZcfoyOp1QzuJmZX4AjAee5XYlHxYuRSmgceVRGFl5FIW0rsYrYritWbEVo8lj2uJuR1Uh0bc5L6UzOj8Ybq5SJNL5RbbSJVatbDbs/PMl/IFBdBx/4vWAkD/F7OBQCDwT3gDovpIfSii+rYAAAAASUVORK5CYII=","orcid":"","institution":"University of Sharjah","correspondingAuthor":true,"prefix":"","firstName":"Iman","middleName":"Akef","lastName":"Khowailed","suffix":""},{"id":623636173,"identity":"5a58e5bf-b659-4610-a0c1-5e2931a0ca69","order_by":1,"name":"Ibrahim M. Moustafa","email":"","orcid":"","institution":"University of Sharjah","correspondingAuthor":false,"prefix":"","firstName":"Ibrahim","middleName":"M.","lastName":"Moustafa","suffix":""},{"id":623636174,"identity":"9c9f5368-e890-46af-b36a-b9000b277aad","order_by":2,"name":"Dalia Ibrahim Hemdan","email":"","orcid":"","institution":"University of Sharjah","correspondingAuthor":false,"prefix":"","firstName":"Dalia","middleName":"Ibrahim","lastName":"Hemdan","suffix":""},{"id":623636175,"identity":"f7accd60-7427-487a-b58b-574af5f00835","order_by":3,"name":"Emre Selman Caniaz","email":"","orcid":"","institution":"Balıkesir University","correspondingAuthor":false,"prefix":"","firstName":"Emre","middleName":"Selman","lastName":"Caniaz","suffix":""},{"id":623636176,"identity":"61a3466c-af64-4508-b654-db22392e74f6","order_by":4,"name":"Cherie Peters-Brinkerhoff","email":"","orcid":"","institution":"West Coast Solutions (United States)","correspondingAuthor":false,"prefix":"","firstName":"Cherie","middleName":"","lastName":"Peters-Brinkerhoff","suffix":""},{"id":623636177,"identity":"0b8803b6-6048-4237-bca8-eefc0a1ade23","order_by":5,"name":"Muneera M. Almurdi","email":"","orcid":"","institution":"King Saud University","correspondingAuthor":false,"prefix":"","firstName":"Muneera","middleName":"M.","lastName":"Almurdi","suffix":""},{"id":623636178,"identity":"c90e6443-efec-4c1f-867f-ca48b6cbdae7","order_by":6,"name":"Abdulrahman M. Alsubiheen","email":"","orcid":"","institution":"King Saud University","correspondingAuthor":false,"prefix":"","firstName":"Abdulrahman","middleName":"M.","lastName":"Alsubiheen","suffix":""}],"badges":[],"createdAt":"2026-01-21 21:38:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8663548/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8663548/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107666968,"identity":"5886dc7f-9a8a-417b-9fbe-7957e8bfb302","added_by":"auto","created_at":"2026-04-23 19:11:57","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":113629,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary (a) and dependence (b) plots for agility (T-test) performance across all participants. Feature attribution demonstrates low and diffuse predictor influence, with no single demographic or postural variable exerting dominant control. This pattern is consistent with the modest predictive performance observed for agility outcomes and reflects the complex, open-skill nature of agility tasks.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/11d589f6aa42397292520843.jpg"},{"id":107706377,"identity":"44bb9774-b65e-49b3-9b52-221eeb488256","added_by":"auto","created_at":"2026-04-24 09:17:58","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":110303,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary (a) and dependence (b) plots for vertical jump height across all participants. Craniovertebral angle (CVA) and body mass index (BMI) emerged as the dominant contributors, indicating that sagittal cervical alignment and body composition strongly influence explosive lower-limb power generation.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/c284f3115da587d482e6f951.jpg"},{"id":107707574,"identity":"da65a37f-6185-4d59-bd9f-1310ebd0c248","added_by":"auto","created_at":"2026-04-24 09:20:37","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":103801,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary (a) and dependence (b) plots for static balance (Stork test). Age and craniovertebral angle demonstrated consistent, monotonic relationships with balance duration, reflecting predominantly linear neuromechanical control mechanisms underlying static postural stability.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/0552ecc3b56e43e5cb61e7ec.jpg"},{"id":107706494,"identity":"7640bef4-8601-45d5-94f7-70be98da968d","added_by":"auto","created_at":"2026-04-24 09:18:11","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":107467,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary (a) and dependence (b) plots for Y-Balance performance. Craniovertebral angle exhibited the strongest contribution, highlighting the role of sagittal alignment in dynamic postural control and reach stability during multi-directional balance tasks.\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/11f83ca9a9a8abff5aa76bb2.jpg"},{"id":107706162,"identity":"1fa5a2ef-94cd-4c6e-a761-78a6f55abde1","added_by":"auto","created_at":"2026-04-24 09:17:33","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":450670,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary plots stratified by COVID exposure status for all performance outcomes. In COVID-exposed athletes, postural and anthropometric predictors showed amplified influence compared with COVID-free participants, suggesting reduced physiological buffering capacity and greater sensitivity to biomechanical inefficiencies following infection.\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/1ba5ef44875c0ccc903b30c8.jpg"},{"id":107666972,"identity":"2b813ab5-a674-40bd-96ff-daabc6923746","added_by":"auto","created_at":"2026-04-23 19:11:57","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":334816,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary plots stratified by head posture classification. Athletes with forward head posture demonstrated greater variability and reduced predictability across balance and agility outcomes, whereas those with normal head posture exhibited more stable and coherent predictor–outcome relationships, indicating more efficient neuromechanical organization.\u003c/p\u003e","description":"","filename":"Picture6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/ab6c2e31fc23f4dad11af36a.jpg"},{"id":108181028,"identity":"2264983f-3e60-46c7-b869-c8287e160624","added_by":"auto","created_at":"2026-04-30 08:56:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1652233,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/17e431d2-ff32-4ddd-a4e8-bc6183de5958.pdf"},{"id":107706205,"identity":"92ac16d2-1b65-4f56-8a52-48f739407959","added_by":"auto","created_at":"2026-04-24 09:17:39","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":47992,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFilesXAIData.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8663548/v1/d6b0241c6f5d574688859147.xlsx"}],"financialInterests":"Competing interest reported. Yes, the authors have competing interests as defined by Nature Research, or other interests that could be perceived to influence the results and/or discussion reported in this paper. These will be disclosed explicitly in the Competing Interests section of the manuscript.\n\nIman","formattedTitle":"The Silent Saboteurs of Athletic Performance: Explainable AI Highlights Spinal Alignment and COVID-19 as Key Determinants","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAthletic performance emerges from the interaction of biological, neuromechanical, demographic, and environmental factors that collectively influence agility, balance, and lower limb power. Conventional predictors such as age, sex, and body mass index (BMI) provide valuable population-level insights; however, these variables account for only a limited proportion of performance variability, particularly within elite or homogeneous athletic cohorts [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. This limitation reflects the fundamentally nonlinear nature of the human motor system, in which neuromuscular drive, postural control, and task execution arise from high-dimensional interactions between central and peripheral mechanisms. Consequently, linear models based primarily on demographic descriptors are insufficient to represent the complex structure underlying athletic performance.\u003c/p\u003e \u003cp\u003eSagittal postural alignment has recently emerged as a critical neuromechanical determinant of motor performance. Forward head posture (FHP), commonly quantified using the craniovertebral angle (CVA), is not merely a structural deviation associated with cervical discomfort. Rather, it reflects altered integration of proprioceptive, vestibular, and visual inputs essential for sensorimotor control [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Clinical evidence demonstrates that correction of sagittal cervical alignment restores proprioceptive feedback, improves nerve root conduction, normalizes autonomic function, and modulates central pain processing [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].Importantly, these effects persist in long-term follow-up studies, supporting the concept that posture functions as a regulator of central nervous system organization rather than a static musculoskeletal parameter.\u003c/p\u003e \u003cp\u003eCervical sagittal alignment varies systematically with demographic factors and lifestyle behaviors, each exerting distinct biomechanical and sensorimotor consequences. Cross-sectional studies show that increasing age and higher BMI are associated with reduced CVA (greater forward head posture), diminished neck muscle endurance, and impaired proprioceptive acuity [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. These demographic effects are further shaped by occupational and behavioral exposures. Prolonged sitting, excessive screen and smartphone use, and inadequate ergonomic environments have been consistently linked to increased forward head posture and neck pain, particularly among office and information technology workers [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. During the COVID-19 pandemic, widespread shifts toward sedentary behavior, remote work, and prolonged screen exposure were associated with a population-level increase in postural deviations and neck-related complaints [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBeyond clinical and occupational settings, sagittal cervical alignment has demonstrated relevance in athletic performance. Our group has shown that cervical sagittal alignment predicts anticipatory core muscle activation, a key prerequisite for explosive movement and rapid change-of-direction tasks [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. In controlled athletic cohorts, forward head posture has been associated with reduced agility, lower limb power deficits, shorter single-leg stance duration, and decreased Y-balance composite scores. These functional impairments were accompanied by altered somatosensory evoked potentials, indicating degraded sensorimotor integration [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Collectively, these findings support the conceptualization of posture as a neuromechanical input variable that shapes the perceptual reference frame from which movement strategies are generated.\u003c/p\u003e \u003cp\u003eIn parallel, systemic health factors introduce additional variability into athletic performance capacity. SARS-CoV-2 exposure represents a biologically plausible but underexplored modifier of neuromuscular performance. Empirical evidence indicates that COVID-19 infection can exert persistent negative effects on balance, movement kinetics, neuromuscular function, and central fatigue, even in young and previously healthy athletic populations [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Given the high prevalence of COVID-19 exposure in contemporary athletes, infection burden may meaningfully influence performance outcomes.\u003c/p\u003e \u003cp\u003eSeveral post\u0026ndash;SARS-CoV-2 sequelae relevant to athletic performance have been identified. Alterations in autonomic regulation, including reduced or dysregulated heart rate variability (HRV), suggest impaired autonomic control that may compromise exercise tolerance and recovery [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Mechanistic studies link mitochondrial dysfunction to long-COVID, providing a biological basis for persistent fatigue and reduced oxidative capacity. Objective assessments demonstrate balance deficits following mild infection, altered running mechanics characterized by prolonged stance time and reduced propulsion, and neuromuscular impairments including prolonged fatigue and reduced power output [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. These findings imply that previously infected athletes may exhibit reduced tolerance to cumulative mechanical stress and require individualized, closely monitored return-to-training strategies.\u003c/p\u003e \u003cp\u003eDespite the growing recognition of posture, demographic factors, and systemic health as determinants of athletic performance, existing predictive models overwhelmingly rely on linear regression or single-domain indicators. Such approaches fail to capture nonlinear interactions between neuromechanical alignment, demographic characteristics, and biological stressors. Machine learning (ML) and artificial intelligence (AI) methods offer a powerful alternative by modeling complex, multidimensional dependencies and have demonstrated superior predictive performance in sport analytics, injury prediction, and talent identification[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, current AI applications in sports science rarely integrate postural biomarkers or COVID-related variables, and few provide interpretable outputs capable of informing clinical or training decisions.\u003c/p\u003e \u003cp\u003eExplainable Artificial Intelligence (XAI) addresses this limitation by enabling transparency and interpretability in ML models. XAI techniques identify influential features, clarify decision pathways, and enhance trust in model predictions [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Feature attribution methods are central to XAI, quantifying the contribution of each input variable to model outputs at both global and individual levels. Among these, SHapley Additive exPlanations (SHAP) applies cooperative game theory to assign contribution values to features based on their marginal impact relative to a baseline prediction [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eXAI has increasingly been adopted in sports science to improve precision and clinical relevance in athlete assessment. Applications using SHAP have identified key physiological, biomechanical, and demographic contributors to performance and injury risk [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. In swimming, XAI analyses revealed that body composition, training load, and lower limb power were dominant performance determinants [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. In soccer, XAI applied to neuromuscular and biomechanical data identified predictors of muscle injury, supporting individualized prevention strategies [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Collectively, these findings demonstrate that XAI bridges the gap between predictive accuracy and practical applicability [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAccordingly, the present study hypothesizes that integrating sagittal cervical alignment, demographic characteristics, and COVID-19 exposure variables within an explainable machine learning framework will improve the prediction of athletic performance outcomes compared with conventional linear models. Furthermore, it is hypothesized that XAI-based feature attribution will identify posture-related and systemic health variables as key contributors to performance variability, providing mechanistically interpretable insights relevant to training optimization and injury risk management.\u003c/p\u003e"},{"header":"Methods","content":"\u003ch3\u003eStudy design and data sources\u003c/h3\u003e\n\u003cp\u003eThis study employed a\u0026nbsp;\u003cstrong\u003eretrospective machine learning (ML) design\u003c/strong\u003e using pooled, individual-level data derived from\u0026nbsp;\u003cstrong\u003ethree previously published investigations\u003c/strong\u003e conducted in collegiate athletes. The contributing datasets examined posture, demographic characteristics, and skill-related athletic performance using standardized field-based protocols.\u003c/p\u003e\n\u003cp\u003eThe first dataset was a\u0026nbsp;\u003cstrong\u003e3D posture–performance study\u003c/strong\u003e, which quantified translational and rotational displacement of spinal segments using \u003cem\u003ePostureScreen® Mobile\u003c/em\u003e and assessed athletic performance through agility, lower-limb power, static balance, dynamic balance, and cardiopulmonary exercise testing. The second dataset comprised a\u0026nbsp;forward head posture (FHP)cohort, which compared athletes with craniovertebral angle (CVA)–defined sagittal cervical misalignment to matched controls with normal alignment and reported outcomes including T-test agility, non-countermovement vertical jump, Stork static balance, and Y-balance performance. The third dataset was a\u0026nbsp;Long COVID athlete cohort, which stratified athletes based on documented SARS-CoV-2 infection history and included standardized assessments of agility, vertical jump, static balance, and Y-balance performance.\u003c/p\u003e\n\u003cp\u003eOnly variables that were\u0026nbsp;conceptually equivalent and consistently available across all three datasets\u0026nbsp;were considered for pooled analysis.\u003c/p\u003e\n\u003ch3\u003eEthical approval\u003c/h3\u003e\n\u003cp\u003eAll studies were approved by the relevant institutional ethics committees in accordance with the Declaration of Helsinki, with written informed consent obtained from all participants (University of Sharjah IRB: REC-22-11-26-S; University of Sharjah IRB: REC-22-06-09-03; Cairo University Research Ethics Committee, Faculty of Physical Therapy: P.T.Rec/012/0050051). The present work involved secondary analysis of fully de-identified data, with no participant contact or new data collection.\u003c/p\u003e\n\u003ch3\u003eData harmonization and analytic sample\u003c/h3\u003e\n\u003cp\u003eTo ensure methodological consistency and avoid selective availability bias, a\u0026nbsp;\u003cstrong\u003ec\u003c/strong\u003e\u003cstrong\u003eommon-variable framework\u003c/strong\u003ewas applied. Variables not uniformly available across all datasets were excluded a priori. Accordingly, neurophysiological measures (e.g., somatosensory evoked potential components), radiographic alignment parameters, and cardiopulmonary variables were not included in the current analysis.\u003c/p\u003e\n\u003cp\u003eAfter harmonization, quality screening, and removal of incomplete records, the final analytic sample comprised\u0026nbsp;\u003cstrong\u003e562 collegiate athletes\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003ch3\u003ePredictor variables\u003c/h3\u003e\n\u003cp\u003eSix predictor variables were incorporated into the machine learning models: age (years), body mass index (BMI; kg/m²), COVID-19 count (number of documented prior SARS-CoV-2 infections), craniovertebral angle (CVA; degrees), sex (binary encoded), and sport category (binary encoded).\u003c/p\u003e\n\u003cp\u003eCVA was treated as a continuous biomechanical variable and also used to derive a posture classification. Forward head posture was defined as\u0026nbsp;\u003cstrong\u003eCVA \u0026lt; 50°\u003c/strong\u003e, consistent with established reliability thresholds, whereas\u0026nbsp;\u003cstrong\u003eCVA ≥ 50°\u003c/strong\u003e indicated non-forward head posture. COVID exposure was determined from documented infection history reported in the original datasets, and COVID count reflected cumulative infection burden where applicable.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eOutcome variables\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eFour standardized field-based indicators of athletic performance were modeled as continuous outcome variables: agility (T-test time, seconds), vertical jump height (cm), static balance (Stork test, seconds), and dynamic balance (Y-Balance composite score, %).\u0026nbsp;These outcomes were selected to represent complementary domains of athletic performance, including neuromuscular coordination, explosive power, postural control, and dynamic balance.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eData preprocessing\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eAll datasets were merged into a single analytic table following variable renaming, unit verification, and range screening. Continuous predictors were examined for implausible values and distributional anomalies. For ML modeling, numeric predictors were scaled using statistics derived\u0026nbsp;\u003cstrong\u003eexclusively from the training data\u003c/strong\u003e, with identical transformations applied to the test set to prevent data leakage. Categorical predictors were encoded using predefined binary schemes applied consistently across cohorts.\u003c/p\u003e\n\u003cp\u003eMissing data handling was prespecified. When sporadic missing values were present, imputation was performed using training-set–derived statistics only.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eMachine learning model development\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eGiven the anticipated nonlinear and interactive relationships among determinants of athletic performance, a comparative machine learning framework was implemented. The models evaluated included regularized linear regression as a baseline reference model, random forest regression, and gradient boosting regression.\u003c/p\u003e\n\u003cp\u003eThe dataset was randomly partitioned into a\u0026nbsp;\u003cstrong\u003etraining set (80%)\u003c/strong\u003eand an\u003cstrong\u003eindependent test set (20%)\u003c/strong\u003e. Hyperparameter tuning was conducted within the training set using\u0026nbsp;\u003cstrong\u003ek-fold cross-validation\u003c/strong\u003e\u003cstrong\u003e.\u0026nbsp;\u003c/strong\u003eModel selection was based on cross-validated performance, and final model evaluation was performed on the held-out test set.\u003c/p\u003e\n\u003cp\u003eModel performance was quantified using the coefficient of determination \u003cstrong\u003e(\u003c/strong\u003e\u003cstrong\u003eR²\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e as the primary metric, supplemented by absolute error measures (e.g., RMSE and MAE) to support interpretability of prediction error in native units.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eExplainable artificial intelligence analysis\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eTo enhance interpretability,\u0026nbsp;\u003cstrong\u003eSHapley Additive exPlanations (SHAP)\u003c/strong\u003e were applied to the final models. Feature attribution was examined at both the global level (mean absolute SHAP values) and the individual level to identify the relative contribution of posture, demographic, and COVID-related variables to each performance outcome. Dependence plots were used to explore linear and nonlinear relationships between predictors and outcomes.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eSubgroup and sensitivity analyses\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003ePrespecified subgroup analyses were conducted to examine whether posture–performance relationships differed by COVID exposure status and posture classification (FHP vs non-FHP). Sensitivity analyses were performed to confirm the stability of predictor importance rankings across subgroups.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eSoftware and reproducibility\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eAll analyses were performed using Python-based machine learning libraries. A fixed random seed was applied to data splitting, model training, and cross-validation procedures to ensure reproducibility. The full preprocessing and modeling pipeline was implemented as a unified workflow to minimize analytic variability across outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data generated and/or analyzed during the current study are included in the published article and its supplementary information files.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Data Preparation and Descriptive Results\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing data cleaning and harmonization procedures, a total of \u003cstrong\u003e562 participants\u003c/strong\u003e were retained for the final analysis. Records containing missing values, implausible entries, or empty variables were excluded to ensure data integrity. All numerical predictors were standardized prior to model training to place variables on comparable scales and to facilitate stable and unbiased machine-learning estimation.\u003c/p\u003e\n\u003cp\u003eDescriptive statistics for demographic, postural, and athletic performance variables are summarized in \u003cstrong\u003eTable 1\u003c/strong\u003e. The cohort represented a relatively homogeneous athletic population, characterized by a narrow age range and body mass index distribution. This homogeneity reduces demographic confounding and allows model performance to more directly reflect neuromechanical and postural influences rather than broad population variability. Craniovertebral angle (CVA) values spanned both forward head posture and normal alignment ranges, providing sufficient variability to support posture-based subgroup analyses and interpretable modeling of posture\u0026ndash;performance relationships.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e Descriptive statistics for demographic, postural, and performance variables.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eMean \u0026plusmn; SD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eMin\u0026ndash;Max\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003en (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e22.46 \u0026plusmn; 1.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e17\u0026ndash;26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBMI (kg/m\u0026sup2;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e22.15 \u0026plusmn; 1.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e18.50\u0026ndash;24.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCOVID_Count\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.04 \u0026plusmn; 0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0\u0026ndash;3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCVA (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e49.76 \u0026plusmn; 2.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e41.96\u0026ndash;57.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAgility_TTest_s\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e9.70 \u0026plusmn; 1.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.00\u0026ndash;13.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eVertical_Jump_cm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e40.44 \u0026plusmn; 1.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e35.21\u0026ndash;44.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eStork_Balance_s\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e23.97 \u0026plusmn; 1.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e20.90\u0026ndash;27.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eY_Balance_%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e85.12 \u0026plusmn; 1.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e81.07\u0026ndash;89.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSex \u0026ndash; Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e283 (50.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSex \u0026ndash; Male\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e279 (49.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSport Category \u0026ndash; Lower Limb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e426 (75.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSport Category \u0026ndash; Upper Limb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e136 (24.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eSubgroup stratification was performed based on \u003cstrong\u003eCOVID exposure status\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;and \u003cstrong\u003epostural alignment\u003c/strong\u003e\u003c/strong\u003e to enable targeted comparative modeling. Sixty-one participants were classified as COVID-free (COVID_Count = 0), whereas 501 participants had a documented history of one or more SARS-CoV-2 infections. Based on established craniovertebral angle thresholds, 291 participants were categorized as exhibiting forward head posture (CVA \u0026lt; 50\u0026deg;), and 271 were classified as having normal head posture (CVA \u0026ge; 50\u0026deg;). Subgroup distributions are presented in \u003cstrong\u003eTable 2\u003c/strong\u003e. These stratifications allowed examination of whether posture\u0026ndash;performance relationships and predictor importance patterns differed according to systemic health status and sagittal alignment, thereby supporting interpretable subgroup-specific machine-learning analyses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Sample distribution for each subgroup.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSubgroup\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDefinition\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003en\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCOVID-Free\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCOVID_Count = 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e61\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCOVID-Exposed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCOVID_Count \u0026ge; 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e501\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eForward Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCVA \u0026lt; 50\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e291\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNormal Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCVA \u0026ge; 50\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e271\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTotal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Model Performance Across Outcomes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThree machine-learning models\u0026mdash;\u003cstrong\u003eLinear Regression, Random Forest, and XGBoost\u003c/strong\u003e\u0026mdash;were trained for each athletic performance outcome to compare linear and nonlinear modeling approaches. Model performance metrics, including the coefficient of determination (R\u0026sup2;), root mean square error (RMSE), and mean absolute error (MAE), are summarized in \u003cstrong\u003eTable 3\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eAcross all outcomes, \u003cstrong\u003eLinear Regression consistently achieved the highest explanatory power and lowest prediction error\u003c/strong\u003e, outperforming tree-based models. This consistent pattern indicates that the dominant relationships linking postural alignment, demographic variables, and athletic performance in this cohort were predominantly linear in nature, with limited evidence of strong nonlinear interactions. These findings suggest that, within a relatively homogeneous athletic population, performance variability is largely governed by proportional biomechanical and physiological relationships rather than complex hierarchical feature interactions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u003c/strong\u003e Model performance (R\u0026sup2;, RMSE, MAE) for all outcomes.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eOutcome\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eR\u0026sup2;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eRMSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eMAE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.346\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.834\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.628\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.193\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.927\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.686\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.089\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.984\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.724\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.893\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.573\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.351\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.627\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.422\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.859\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.656\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.439\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.767\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.704\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.471\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.528\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.711\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.567\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.869\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.339\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.843\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.624\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.394\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.649\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.435\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.1 Agility (T-Test; lower is better)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAgility performance demonstrated the weakest predictability across all models. Linear Regression achieved an R\u0026sup2; of 0.35 with an RMSE of 0.83, whereas Random Forest and XGBoost showed lower explanatory power (R\u0026sup2; = 0.19 and 0.08, respectively) and higher prediction error. SHAP-based explainability results for agility are presented in \u003cstrong\u003eFigure 1a\u0026ndash;b\u003c/strong\u003e. Predictor contributions were relatively diffuse, with no single anthropometric or postural variable exerting dominant influence. This pattern suggests that agility performance is only partially explained by static structural characteristics and likely reflects a greater dependence on higher-order neuromotor, perceptual, and decision-making processes that are not captured by the current input features.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 1.\u003c/strong\u003e SHAP summary (a) and dependence (b) plots for agility (T-test) performance across all participants. Feature attribution demonstrates low and diffuse predictor influence, with no single demographic or postural variable exerting dominant control. This pattern is consistent with the modest predictive performance observed for agility outcomes and reflects the complex, open-skill nature of agility tasks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.2 Vertical Jump (higher is better)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eVertical jump height exhibited the strongest predictability among all outcomes. Linear Regression achieved the highest coefficient of determination (R\u0026sup2; = 0.89) with the lowest RMSE (0.57), followed by Random Forest (R\u0026sup2; = 0.87, RMSE = 0.62) and XGBoost (R\u0026sup2; = 0.86, RMSE = 0.65). SHAP plots for this outcome are shown in \u003cstrong\u003eFigure 2a\u0026ndash;b\u003c/strong\u003e, demonstrating concentrated feature attribution. The high predictive accuracy and focused SHAP contributions indicate that vertical jump performance is strongly governed by proportional biomechanical relationships linking sagittal alignment and body composition to lower-limb power generation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 2.\u003c/strong\u003e SHAP summary (a) and dependence (b) plots for vertical jump height across all participants. Craniovertebral angle (CVA) and body mass index (BMI) emerged as the dominant contributors, indicating that sagittal cervical alignment and body composition strongly influence explosive lower-limb power generation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.3 Stork Balance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor static balance performance, Linear Regression achieved an R\u0026sup2; of 0.77 with an RMSE of 0.70, outperforming Random Forest and XGBoost (both R\u0026sup2; \u0026asymp; 0.71, RMSE \u0026asymp; 0.78). SHAP plots for Stork balance are presented in \u003cstrong\u003eFigure 3a\u0026ndash;b\u003c/strong\u003e. Balance performance showed strong linear relationships, particularly with age and craniovertebral angle. The monotonic nature of these relationships indicates that static postural stability is primarily governed by proportional neuromechanical control mechanisms, with limited contribution from higher-order nonlinear interactions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 3.\u003c/strong\u003e SHAP summary (a) and dependence (b) plots for static balance (Stork test). Age and craniovertebral angle demonstrated consistent, monotonic relationships with balance duration, reflecting predominantly linear neuromechanical control mechanisms underlying static postural stability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.4 Y-Balance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eY-Balance performance was strongly predictable across all models. Linear Regression achieved the highest R\u0026sup2; (0.87) with the lowest RMSE (0.56), compared with Random Forest (R\u0026sup2; = 0.84, RMSE = 0.62) and XGBoost (R\u0026sup2; = 0.83, RMSE = 0.64). SHAP plots for this outcome are shown in \u003cstrong\u003eFigure 4a\u0026ndash;b\u003c/strong\u003e, indicating substantial contributions from anthropometric and postural indicators. The concentrated feature attribution and strong linear model performance suggest that dynamic balance during multi-directional reach tasks is closely linked to sagittal alignment and body composition, reflecting structured neuromechanical control rather than highly adaptive or nonlinear motor strategies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 4.\u003c/strong\u003e SHAP summary (a) and dependence (b) plots for Y-Balance performance. Craniovertebral angle exhibited the strongest contribution, highlighting the role of sagittal alignment in dynamic postural control and reach stability during multi-directional balance tasks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Subgroup Modeling\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSeparate models were developed for each predefined subgroup to evaluate whether predictor\u0026ndash;outcome relationships differed according to COVID exposure status and postural alignment. Performance comparisons across subgroups are summarized in \u003cstrong\u003eTable 4\u003c/strong\u003e. Across all subgroups and outcomes, \u003cstrong\u003eLinear Regression consistently remained the best-performing model\u003c/strong\u003e, indicating that subgroup stratification primarily influenced model stability and explanatory strength rather than altering the underlying structure of predictor\u0026ndash;outcome relationships.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4.\u003c/strong\u003e Performance comparisons across subgroups.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eSubgroup\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutcome\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eBest R\u0026sup2;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eBest Model\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Free\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Free\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Free\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Free\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Exposed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Exposed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Exposed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eCOVID-Exposed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eForward Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eForward Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eForward Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eForward Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eNormal Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eAgility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eNormal Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eVertical Jump\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eNormal Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eStork Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eNormal Head\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eY-Balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eTable 4. Performance comparisons across subgroups.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3.1 COVID-Free vs COVID-Exposed\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCOVID-free participants demonstrated weaker model fits across all outcomes, reflected by lower R\u0026sup2; values. This pattern likely reflects increased internal variability and reduced signal-to-noise ratio within the smaller COVID-free subgroup (n = 61), rather than a fundamental change in predictor relevance. In contrast, COVID-exposed participants exhibited more stable and consistent model performance across outcomes, suggesting that athletic performance in this group was more tightly constrained by measurable postural and anthropometric factors. SHAP-based subgroup explainability results are presented in \u003cstrong\u003eFigure 5a\u0026ndash;d\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 5.\u003c/strong\u003e SHAP summary plots stratified by COVID exposure status for all performance outcomes. In COVID-exposed athletes, postural and anthropometric predictors showed amplified influence compared with COVID-free participants, suggesting reduced physiological buffering capacity and greater sensitivity to biomechanical inefficiencies following infection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3.2 Forward Head vs Normal Head Posture\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants with forward head posture demonstrated lower predictability for balance and agility outcomes, indicating greater internal variability and disrupted neuromechanical organization within this subgroup. In contrast, the normal head posture subgroup exhibited higher R\u0026sup2; values and more consistent biomechanical patterns across outcomes. These findings suggest that preserved sagittal alignment is associated with more stable and coherent posture\u0026ndash;performance relationships. Posture-based SHAP comparisons are visualized in \u003cstrong\u003eFigure 6a\u0026ndash;d\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 6.\u003c/strong\u003e SHAP summary plots stratified by head posture classification. Athletes with forward head posture demonstrated greater variability and reduced predictability across balance and agility outcomes, whereas those with normal head posture exhibited more stable and coherent predictor\u0026ndash;outcome relationships, indicating more efficient neuromechanical organization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Explainability Analysis (SHAP Results)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4.1 Global Findings (Figures 1\u0026ndash;4)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAcross all models and athletic performance outcomes, SHAP analysis revealed a \u003cstrong\u003ehighly consistent hierarchy of predictor importance\u003c/strong\u003e, indicating robust and stable feature\u0026ndash;outcome relationships. Craniovertebral angle (CVA) consistently emerged as one of the most influential predictors, particularly for balance-related outcomes and vertical jump performance, underscoring the central role of sagittal cervical alignment in neuromechanical control and force transmission. Body mass index (BMI) demonstrated a stable and meaningful contribution to agility and vertical jump outcomes, reflecting the influence of body composition on movement efficiency and power generation. Age showed a clear negative association with balance performance, consistent with age-related changes in postural control mechanisms. COVID_Count exhibited measurable but comparatively smaller effects overall, with its strongest influence observed for agility performance, suggesting sensitivity of open-skill tasks to systemic health perturbations. Global SHAP summary and bar plots illustrating these relationships are presented in \u003cstrong\u003eFigures 1\u0026ndash;4\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4.2 Subgroup SHAP Patterns (Figures 5\u0026ndash;6)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSubgroup-specific SHAP analyses revealed \u003cstrong\u003esystematic shifts in predictor dominance\u003c/strong\u003e, indicating that systemic health status and postural alignment modulate the relative contribution of performance determinants rather than introducing new predictors. In COVID-free participants, age emerged as the dominant contributor across outcomes, suggesting that in the absence of infection-related stressors, performance variability is primarily governed by baseline physiological maturation and neuromuscular efficiency. In contrast, among COVID-exposed participants, CVA and BMI gained greater relative importance, indicating increased reliance on structural alignment and anthropometric factors when physiological buffering capacity may be reduced.\u003c/p\u003e\n\u003cp\u003ePosture-based subgroup analyses further demonstrated that within the forward head posture subgroup, CVA exerted the strongest negative contribution to balance outcomes, reflecting disrupted neuromechanical organization and reduced postural stability. Conversely, in the normal head posture subgroup, predictor influences were more evenly distributed, consistent with a more integrated and resilient neuromotor control system. Subgroup SHAP visualizations are shown in \u003cstrong\u003eFigures 5 and 6\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Summary of Key Findings\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eVertical jump and Y-Balance outcomes demonstrated strong model performance (R\u0026sup2; \u0026ge; 0.86), indicating that these performance domains are highly predictable from postural and anthropometric features. In contrast, agility exhibited weaker predictability, suggesting that additional neuromotor, perceptual, or cognitive variables may be required to fully capture the determinants of open-skill athletic tasks. Explainability analysis consistently identified \u003cstrong\u003ecraniovertebral angle, body mass index, and age\u003c/strong\u003e as the most influential predictors across outcomes. Subgroup analyses further demonstrated that \u003cstrong\u003eCOVID history and head posture modulate the strength and stability of these relationships\u003c/strong\u003e, highlighting the importance of integrating both systemic health status and sagittal alignment when interpreting athletic performance.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study employed interpretable machine-learning models to investigate how sagittal cervical alignment, demographic factors, and COVID-19 exposure history jointly relate to athletic performance in a homogeneous cohort of collegiate athletes. By integrating individual-level data from multiple previously published investigations, the present analysis extends prior work by providing a unified predictive and explainable framework linking posture-related variables to balance, power, and agility outcomes. The principal findings indicate that craniovertebral angle (CVA), body mass index (BMI), and age are consistently associated with balance and power-related performance, whereas agility performance is less strongly explained by static anthropometric and postural features.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eModel Behavior and Predictive Structure\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eAcross all performance outcomes, linear regression consistently outperformed Random Forest and XGBoost models, yielding higher explained variance and lower prediction error. This finding suggests that, within this relatively homogeneous athletic cohort, the dominant relationships between posture, demographic variables, and athletic performance are primarily linear. Although neuromuscular control is inherently nonlinear, the current results indicate that such nonlinearities were either weak, masked by measurement noise, or not captured by the available feature set. Similar observations have been reported in athletic and biomechanical datasets characterized by restricted demographic variability and limited feature dimensionality, where linear models often outperform more complex approaches[22,23].\u003c/p\u003e\n\u003cp\u003eImportantly, the superior performance of linear models does not imply the absence of nonlinear physiological mechanisms. Rather, it suggests that proportional biomechanical relationships—such as those linking sagittal alignment to force transmission and postural stability—account for a substantial proportion of observable performance variability in this cohort. The absence of high-dimensional neuromotor, cognitive, or training-load variables likely constrained the capacity of tree-based models to exploit nonlinear interactions.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eOutcome-Specific Interpretation\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eVertical jump and balance-related outcomes demonstrated strong predictability, with coefficients of determination exceeding 0.77 across models. Explainability analysis revealed that CVA and BMI were dominant contributors to vertical jump performance, supporting the notion that sagittal cervical alignment and body composition influence lower-limb power generation through biomechanical coupling and postural organization\u0026nbsp;[24–26]. These findings align with prior experimental and clinical studies demonstrating that altered cervical alignment is associated with impaired neuromuscular coordination and reduced force efficiency\u0026nbsp;[4,27].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStatic and dynamic balance outcomes (Stork balance and Y-Balance) were also strongly associated with CVA and age. The monotonic relationships observed in SHAP dependence plots suggest that balance performance is governed by proportional neuromechanical control mechanisms rather than highly adaptive or nonlinear strategies. This observation is consistent with previous work linking sagittal alignment to sensorimotor integration, proprioceptive accuracy, and postural stability[28–30].\u003c/p\u003e\n\u003cp\u003eIn contrast, agility performance exhibited substantially weaker predictability across all models. SHAP attribution patterns were diffuse, with no single predictor exerting dominant influence. Agility tasks are inherently open skill in nature, requiring rapid perceptual processing, anticipatory control, and decision-making under time constraints. These higher order neuromotor and cognitive components are not captured by static anthropometric or postural measures, which likely explains the limited explanatory power observed in the present models. Similar limitations have been reported in prior performance modeling studies, where agility outcomes required task-specific or neurocognitive inputs to achieve meaningful prediction accuracy[31,32].\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eSubgroup-Specific Effects of COVID Exposure and Postural Alignment\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eSubgroup analyses revealed systematic differences in model stability and predictor importance without altering the fundamental structure of predictor–outcome relationships. COVID-free athletes demonstrated weaker model fits across outcomes, likely reflecting increased internal variability and reduced statistical power associated with the smaller subgroup size. In contrast, COVID-exposed athletes exhibited more stable and consistent predictive relationships, with CVA and BMI assuming greater relative importance. These findings suggest that following systemic stressors such as SARS-CoV-2 infection, athletic performance may become more tightly constrained by measurable biomechanical and anthropometric factors, potentially reflecting reduced physiological buffering capacity[12,33,34].\u003c/p\u003e\n\u003cp\u003ePosture-based stratification further highlighted meaningful differences. Athletes with forward head posture showed lower predictability for balance and agility outcomes, indicating greater internal variability and disrupted neuromechanical organization. Conversely, athletes with normal head posture demonstrated higher explained variance and more evenly distributed predictor contributions, consistent with more integrated and resilient postural control systems. These observations align with previous reports linking forward head posture to altered sensorimotor integration and compromised postural stability[28,29,35].\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eExplainability and Clinical Interpretability\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eA key strength of this study lies in the application of SHAP-based explainability methods, which enabled transparent interpretation of model predictions at both global and subgroup levels. Across all outcomes, CVA consistently emerged as one of the most influential predictors, underscoring the relevance of sagittal cervical alignment as a neuromechanical factor rather than a purely structural descriptor. BMI contributed meaningfully to power and agility outcomes, while age showed a consistent negative association with balance performance, reflecting well-established age-related changes in postural control\u0026nbsp;[36].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCOVID_Count demonstrated smaller but detectable effects, with its strongest influence observed in agility outcomes. This pattern suggests that open-skill tasks may be particularly sensitive to subtle systemic or neuromotor perturbations following infection, even in young athletic populations. Importantly, subgroup SHAP analyses revealed shifts in predictor dominance rather than the emergence of new predictors, indicating that systemic health status and postural alignment modulate the relative importance of performance determinants rather than redefining them.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003eThis research was funded by the Ongoing Research Funding program (ORF-2026-798), King Saud University, Riyadh, Saudi Arabia.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eLimitations\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eSeveral limitations should be acknowledged. First, the retrospective and observational design precludes causal inference; SHAP-derived feature importance reflects association rather than mechanistic causality. Second, COVID exposure was quantified using a simplified count variable, without information on infection severity, symptom persistence, vaccination status, or time since infection. Third, pooling data from multiple studies introduces potential protocol heterogeneity despite harmonization efforts. Fourth, the absence of neuromotor, cognitive, training-load, and injury-history variables likely limited predictive performance, particularly for agility outcomes. Finally, external validation in an independent cohort is required to confirm generalizability.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eImplications and Future Directions\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eDespite these limitations, the present findings demonstrate that sagittal cervical alignment and basic anthropometric characteristics are robustly associated with balance and power-related athletic performance. The results support the potential inclusion of posture-based metrics within performance screening and monitoring frameworks, while emphasizing that such measures should be interpreted as contributory rather than determinative. Future research should incorporate prospective designs, richer neuromotor feature sets, and injury outcomes to evaluate whether posture-informed, explainable machine-learning models can meaningfully inform performance optimization and injury risk stratification.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, sagittal cervical alignment and basic anthropometric factors are strongly associated with balance and power-related athletic performance, whereas agility is less well explained by static postural and demographic variables. Explainable machine-learning analysis consistently identified craniovertebral angle as a key contributor to balance and vertical jump outcomes, with smaller but detectable effects of COVID-19 exposure history. These findings highlight the value of posture-based metrics within interpretable modeling frameworks and underscore the importance of considering both structural alignment and systemic health status when evaluating athletic performance.\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003cp\u003eYes, the authors have competing interests as defined by Nature Research, or other interests that could be perceived to influence the results and/or discussion reported in this paper. These will be disclosed explicitly in the Competing Interests section of the manuscript.Iman\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis research was funded by the Ongoing Research Funding program (ORF-2026-798), King Saud University, Riyadh, Saudi Arabia.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eI.K. and I.M. conceived and designed the study. Data curation and methodological development were performed by I.K., I.M., D.M., C.P., and E.C. Machine learning modeling and explainable AI analyses were conducted by I.M., I.K., E.C. Figures and data visualization were prepared by I.K. and I.M. Manuscript drafting and writing were carried out by I.K., D.M., C.P., I.M. and A.A. contributed to interpretation of the results and critical revision of the manuscript. All authors reviewed, edited, and approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll data generated and/or analyzed during the current study are included in the published article and its supplementary information files.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBehm, D. G., Muehlbauer, T., Kibele, A. \u0026amp; Granacher, U. Effects of Strength Training Using Unstable Surfaces on Strength, Power and Balance Performance Across the Lifespan: A Systematic Review and Meta-analysis. \u003cem\u003eSports Med.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 1645\u0026ndash;1669 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePanjabi, M. M. The Stabilizing System of the Spine. Part I. Function, Dysfunction, Adaptation, and Enhancement. \u003cem\u003eJ. Spinal Disord\u003c/em\u003e. \u003cb\u003e5\u003c/b\u003e, 383\u0026ndash;389 (1992).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoustafa, I. M., Diab, A. A., Hegazy, F. \u0026amp; Harrison, D. E. Does improvement towards a normal cervical sagittal configuration aid in the management of cervical myofascial pain syndrome: a 1- year randomized controlled trial. \u003cem\u003eBMC Musculoskelet. Disord\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 396 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoustafa, I. M., Diab, A. A. M., Hegazy, F. A. \u0026amp; Harrison, D. E. Does rehabilitation of cervical lordosis influence sagittal cervical spine flexion extension kinematics in cervical spondylotic radiculopathy subjects? \u003cem\u003eJ. Back Musculoskelet. Rehabil\u003c/em\u003e. \u003cb\u003e30\u003c/b\u003e, 937\u0026ndash;941 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKocur, P. et al. Relationship between age, BMI, head posture and superficial neck muscle stiffness and elasticity in adult women. \u003cem\u003eSci Rep\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStincel, O. R. et al. Assessment of Forward Head Posture and Ergonomics in Young IT Professionals \u0026ndash; Reasons to Worry? \u003cem\u003eMedicina del. Lavoro\u003c/em\u003e \u003cb\u003e114\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGomez, I. N., Suarez, C. G., Sosa, K. E. \u0026amp; Tapang, M. L. Work from home-related musculoskeletal pain during the COVID-19 pandemic: A rapid review. \u003cem\u003eInternational Journal of Osteopathic Medicine\u003c/em\u003e vol. 47 Preprint at (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijosm.2022.12.001\u003c/span\u003e\u003cspan address=\"10.1016/j.ijosm.2022.12.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoustafa, I. M., Diab, A. A. \u0026amp; Harrison, D. E. The Efficacy of Cervical Lordosis Rehabilitation for Nerve Root Function and Pain in Cervical Spondylotic Radiculopathy: A Randomized Trial with 2-Year Follow-Up. \u003cem\u003eJ. Clin. Med.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 6515 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoustafa, I., Kim, M. \u0026amp; Harrison, D. E. Comparison of Sensorimotor Integration and Skill-Related Physical Fitness Components Between College Athletes With and Without Forward Head Posture. \u003cem\u003eJ. Sport Rehabil\u003c/em\u003e. \u003cb\u003e32\u003c/b\u003e, 53\u0026ndash;62 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJafarnezhadgero, A. A., Noroozi, R., Fakhri, E., Granacher, U. \u0026amp; Oliveira, A. S. The Impact of COVID-19 and Muscle Fatigue on Cardiorespiratory Fitness and Running Kinetics in Female Recreational Runners. \u003cem\u003eFront Physiol\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSedaghati, P., Balayi, E. \u0026amp; Ahmadabadi, S. Effects of COVID-19 related physical inactivity on motor skills in children with intellectual disability. \u003cem\u003eBMC Public. Health\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZacher, J., Branahl, A., Predel, H. G. \u0026amp; Laborde, S. Effects of Covid-19 on the autonomic nervous system in elite athletes assessed by heart rate variability. \u003cem\u003eSport Sci. Health\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 1269\u0026ndash;1280 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMolnar, T. et al. Mitochondrial dysfunction in long COVID: mechanisms, consequences, and potential therapeutic approaches. \u003cem\u003eGeroScience\u003c/em\u003e vol. 46 5267\u0026ndash;5286 Preprint at (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11357-024-01165-5\u003c/span\u003e\u003cspan address=\"10.1007/s11357-024-01165-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaborri, J. et al. A Machine-Learning Approach to Measure the Anterior Cruciate Ligament Injury Risk in Female Basketball Players. \u003cem\u003eSensors\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e, 3141 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePietraszewski, P. et al. The Role of Artificial Intelligence in Sports Analytics: A Systematic Review and Meta-Analysis of Performance Trends. \u003cem\u003eAppl. Sci.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 7254 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarredo Arrieta, A. et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. \u003cem\u003eInform. Fusion\u003c/em\u003e. \u003cb\u003e58\u003c/b\u003e, 82\u0026ndash;115 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLinardatos, P., Papastefanopoulos, V. \u0026amp; Kotsiantis, S. Explainable AI: A Review of Machine Learning Interpretability Methods. \u003cem\u003eEntropy\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 18 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKranzinger, S., Halmich, C., Hofer, D. \u0026amp; Kranzinger, C. A scoping review of explainable artificial intelligence in sports science. \u003cem\u003eDiscover Artif. Intell.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s44163-025-00709-8\u003c/span\u003e\u003cspan address=\"10.1007/s44163-025-00709-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarvalho, D. D. et al. Swimming Performance Interpreted through Explainable Artificial Intelligence (XAI)\u0026mdash;Practical Tests and Training Variables Modelling. \u003cem\u003eApplied Sci. (Switzerland)\u003c/em\u003e 14, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCalder\u0026oacute;n-D\u0026iacute;az, M. et al. Explainable Machine Learning Techniques to Predict Muscle Injuries in Professional Soccer Players through Biomechanical Analysis. \u003cem\u003eSensors\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSouaifi, M. et al. Artificial Intelligence in Sports Biomechanics: A Scoping Review on Wearable Technology, Motion Analysis, and Injury Prevention. \u003cem\u003eBioengineering\u003c/em\u003e vol. 12 Preprint at (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/bioengineering12080887\u003c/span\u003e\u003cspan address=\"10.3390/bioengineering12080887\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeckey, C., Van Dyk, N., Doherty, C., Lawlor, A. \u0026amp; Delahunt, E. Machine learning approaches to injury risk prediction in sport: A scoping review with evidence synthesis. \u003cem\u003eBritish Journal of Sports Medicine\u003c/em\u003e vol. 59 491\u0026ndash;500 Preprint at (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bjsports-2024-108576\u003c/span\u003e\u003cspan address=\"10.1136/bjsports-2024-108576\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJauhiainen, S. et al. Predicting ACL Injury Using Machine Learning on Data From an Extensive Screening Test Battery of 880 Female Elite Athletes. \u003cem\u003eAm. J. Sports Med.\u003c/em\u003e \u003cb\u003e50\u003c/b\u003e, 2917\u0026ndash;2924 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuu, B. C. et al. Machine Learning Outperforms Logistic Regression Analysis to Predict Next-Season NHL Player Injury: An Analysis of 2322 Players From 2007 to 2017. \u003cem\u003eOrthop J. Sports Med\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarkovic, G. \u0026amp; Jaric, S. Is vertical jump height a body size-independent measure of muscle power? \u003cem\u003eJ. Sports Sci.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e, 1355\u0026ndash;1363 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBen Mansour, G., Kacem, A., Ishak, M., Gr\u0026eacute;lot, L. \u0026amp; Ftaiti, F. The effect of body composition on strength and power in male and female students. \u003cem\u003eBMC Sports Sci. Med. Rehabil\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTamim, M., Moustafa, I. M., Alaparthi, G. K., Oakley, P. A. \u0026amp; Harrison, D. E. Translational and Rotational Postural Aberrations Are Related to Pulmonary Functions and Skill-Related Physical Fitness Components in Collegiate Athletes. \u003cem\u003eJ Clin. Med\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarrison, D. D. et al. Modeling of the Sagittal Cervical Spine as a Method to Discriminate Hypolordosis. \u003cem\u003eSpine (Phila Pa. 1976)\u003c/em\u003e. \u003cb\u003e29\u003c/b\u003e, 2485\u0026ndash;2492 (2004).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTreleaven, J. Sensorimotor disturbances in neck disorders affecting postural stability, head and eye movement control. \u003cem\u003eMan. Ther.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 2\u0026ndash;11 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHa, S. Y. \u0026amp; Sung, Y. H. A temporary forward head posture decreases function of cervical proprioception. \u003cem\u003eJ. Exerc. Rehabil\u003c/em\u003e. \u003cb\u003e16\u003c/b\u003e, 168\u0026ndash;174 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin, G., Zhao, X., Wang, W. \u0026amp; Wilkinson, T. The relationship between forward head posture, postural control and gait: A systematic review. \u003cem\u003eGait and Posture\u003c/em\u003e vol. 98 316\u0026ndash;329 Preprint at (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.gaitpost.2022.10.008\u003c/span\u003e\u003cspan address=\"10.1016/j.gaitpost.2022.10.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheppard, J. M. \u0026amp; Young, W. B. Agility literature review: Classifications, training and testing. \u003cem\u003eJ. Sports Sci.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 919\u0026ndash;932 (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScanlan, A., Humphries, B., Tucker, P. S. \u0026amp; Dalbo, V. The influence of physical and cognitive factors on reactive agility performance in men basketball players. \u003cem\u003eJ. Sports Sci.\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, 367\u0026ndash;374 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuzik, A., Wolan-Nieroda, A., Kochman, M., Perenc, L. \u0026amp; Drużbicki, M. Impact of mild COVID-19 on balance function in young adults, a prospective observational study. \u003cem\u003eSci Rep\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMolnar, T. et al. Mitochondrial dysfunction in long COVID: mechanisms, consequences, and potential therapeutic approaches. \u003cem\u003eGeroScience\u003c/em\u003e vol. 46 5267\u0026ndash;5286 Preprint at (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11357-024-01165-5\u003c/span\u003e\u003cspan address=\"10.1007/s11357-024-01165-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXia, Q. et al. Factors associated with balance impairments in the community-dwelling elderly in urban China. \u003cem\u003eBMC Geriatr\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Athletic Performance, Postural Balance, Machine Learning, COVID-19","lastPublishedDoi":"10.21203/rs.3.rs-8663548/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8663548/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAthletic performance arises from the interaction of biomechanical alignment, demographic characteristics, and systemic health status; however, the relative contribution of these factors remains incompletely understood. This study investigated whether sagittal cervical alignment, demographic variables, and COVID-19 exposure history jointly predict athletic performance outcomes using an explainable machine-learning framework. A retrospective analysis was conducted on pooled individual-level data from three previously published studies, comprising 562 collegiate athletes. Six predictors age, body mass index (BMI), sex, sport category, craniovertebral angle (CVA), and COVID infection count were used to model four standardized performance outcomes: agility (T-test), vertical jump height, static balance (Stork test), and dynamic balance (Y-Balance). Linear regression, random forest, and gradient boosting models were trained and evaluated, with explainability assessed using SHapley Additive exPlanations (SHAP). Across all outcomes, linear regression consistently outperformed tree-based models, indicating predominantly linear relationships between predictors and performance measures. Vertical jump and Y-Balance outcomes demonstrated strong predictability (R\u0026sup2; \u0026ge; 0.86), whereas agility showed weaker model performance (R\u0026sup2; \u0026le; 0.35). Explainability analysis identified CVA as a dominant predictor of balance and power outcomes, while BMI and age contributed meaningfully to performance variability. COVID infection history exerted smaller but detectable effects, particularly for agility. Subgroup analyses revealed that COVID-exposed athletes and those with forward head posture relied more on postural and anthropometric predictors, whereas athletes with normal head posture demonstrated more stable predictor\u0026ndash;outcome relationships. These findings indicate that sagittal cervical alignment is a robust neuromechanical contributor to balance and power-related athletic performance, while agility depends more strongly on unmeasured neuromotor and perceptual factors. Integrating posture-based metrics within explainable machine-learning models provides interpretable insights into performance variability and highlights the importance of considering both structural alignment and systemic health status in athlete assessment.\u003c/p\u003e","manuscriptTitle":"The Silent Saboteurs of Athletic Performance: Explainable AI Highlights Spinal Alignment and COVID-19 as Key Determinants","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-23 19:11:53","doi":"10.21203/rs.3.rs-8663548/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-04-15T10:29:14+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-31T19:01:05+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-06T03:49:52+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-04T10:48:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-02-04T10:14:38+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4f4fec73-6548-4e82-ba2a-e69b1354a148","owner":[],"postedDate":"April 23rd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":66371676,"name":"Health sciences/Health care"},{"id":66371677,"name":"Health sciences/Medical research"},{"id":66371678,"name":"Biological sciences/Neuroscience"}],"tags":[],"updatedAt":"2026-04-23T19:11:53+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-23 19:11:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8663548","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8663548","identity":"rs-8663548","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}