Richard W. Willy

Purpose: This study evaluated five publicly available large language model‐based systems using a multidimensional, expert‐based assessment of consensus‐based clinical questions on bone stress injuries. Methods: Ten standardised questions derived from a 2025 international Delphi consensus on bone stress injuries were submitted to ChatGPT‐5, Perplexity AI, Claude 4, Google Gemini 2.5 and Microsoft Copilot. Responses were independently rated by clinical experts using a multidimensional evaluation framework covering accuracy, comprehensiveness, understanding, reasoning, clarity, potential harm and trust, each rated on a 5‐point Likert scale. Median scores were calculated for each question and model, and comparisons were performed using Kruskal–Wallis tests followed by Dunn's post hoc tests with Bonferroni correction. Interrater agreement was assessed using the weighted Gwet's agreement coefficient. Reproducibility was assessed by repeatedly submitting nine questions representing no, partial, and full expert consensus to each model, with response consistency scored from 1 to 3. Results: Significant differences between models were observed for comprehensiveness, reasoning, clarity and trust (all p < 0.001), with no differences in accuracy or harm. ChatGPT‐5 and Claude 4 achieved significantly higher comprehensiveness scores, while ChatGPT‐5 and Google Gemini 2.5 had significantly higher reasoning scores than Microsoft Copilot. Claude 4 demonstrated significantly lower clarity than all other models. ChatGPT‐5 achieved significantly higher trust scores and showed the highest mean reproducibility score, followed by Perplexity AI and Gemini 2.5; Microsoft Copilot showed the lowest reproducibility. Interrater agreement ranged from fair to almost perfect, with substantial overall agreement. Conclusions: Publicly available large language model‐based systems demonstrated moderate‐to‐high performance in addressing expert‐derived, consensus‐based questions on bone stress injuries, suggesting potential to support the communication of consensus‐based knowledge. However, performance varied across evaluation dimensions and reproducibility, indicating that large language models may support, but should not replace, primary consensus documents or expert clinical judgement. Level of Evidence: N/A.

We examined the influence of running cadence manipulation on femoral cartilage deformation in runners and assessed the association between knee biomechanics and cartilage characteristics. Thirty‐one runners (15 females and 16 males) completed a habitual and high cadence (+10%) run 2–7 days apart in a counterbalanced order. Participants underwent ultrasound imaging of their femoral trochlear cartilage before and after running. A force instrumented treadmill with an 8‐camera motion capture system obtained the peak knee flexion angle, peak knee extensor moment, peak patellofemoral joint force (PFJF), peak patellofemoral joint stress (PFJS), and cumulative PFJS. A 2 (condition) by 2 (time) ANOVA with repeated measures revealed medial cartilage thickness decreased similarly between conditions after running (1.97 (95% CI: 1.86, 2.09) vs. 1.89 (95% CI: 1.78, 2.01) mm; mean difference = −0.078 (95% CI: −0.130, −0.026) mm, p = 0.005). No main effect of condition or interaction effect was found despite a difference between biomechanics outcomes (all p < 0.001). Higher peak and cumulative PFJS were associated with a greater decrease in medial cartilage thickness following high cadence running ( p = 0.011). Higher PFJS was associated with thicker medial, central, and lateral cartilage thickness (all p < 0.05). Increasing running cadence may be a viable intervention to decrease external joint loading while maintaining an internal stimulus to knee cartilage. Cartilage is avascular and relies on mechanical loading to facilitate nutrient diffusion. As such, joint stress during running may be beneficial for cartilage development.

We validated a wearable accelerometer and running watch for estimating peak vertical ground reaction force (vGRF) and cumulative weighted damage (CWD). First, 18 well‐trained runners of equal sex distribution (3.3 ± 0.3 m/s habitual training pace) ran a range of speeds (2.7–4.0 m/s, plus training pace) on an instrumented treadmill as reference standard vGRF and wearable data were collected. Wearable data were inputted into a published equation to estimate peak vGRF. The wearable and equation systematically underestimated peak vGRF compared with the instrumented treadmill (mean bias: −0.22 BWs with LOA: −0.44 to 0.01; 9.4–10.1%RMSE). A linear regression model was used to derive a corrected equation to improve the estimate of peak vGRF. In the validation stage, the wearable and corrected equation were validated in 10 runners in an independent sample of equal sex distribution (3.2 ± 0.3 m/s habitual training pace) yielding excellent relationships and trivial errors for peak vGRF versus the reference standard (ICC _3,1 = 0.87–0.97; p < 0.001; mean bias −0.008 BWs with LOA: −0.15 to 0.14 BWs; 1.5%–4.0% RMSE). To illustrate how these methods can be applied to runners at risk for a bone stress injury, CWD per km was estimated, resulting in trivial errors by the wearable versus the instrumented treadmill (ICC _3,1 = 0.85–0.97; p < 0.001; 1.5–4.1%RMSE). In summary, a commercial wearable and corrected equation were valid estimates of peak vGRF and CWD. These methods have potential for monitoring training loads in runners at risk for bone stress injuries. An online calculator is provided to estimate peak vGRF and CWD from the running watch and wearable.

Context : Running is good for overall health but has a poorly understood risk of injury. Wearable technology and machine learning (ML) offer solutions to this problem. Using ML, we aimed to determine (1) the key important factors that increase the probability of classifying a run as belonging to an injured group, (2) the best-performing ML model, and (3) the sample size required for future substantive studies. Design : Exploratory classification analysis. Methods : Heart rate/inertial measurement unit/GPS data were extracted from the wristwatches of 88 recreational runners (65 injured and 23 uninjured) during a 12-week prospective cohort. Data were extracted from each run completed by participants, and each run was considered an independent event. Variance inflation factor was used to check for collinearity, and the robust normalizer was used to scale numerical features. Seven ML models were used and validated using repeated stratified 10-fold cross validation. Accuracy, precision, and recall were used to evaluate ML model performance and SHapley Additive exPlanations to determine the relative importance of the individual factors. Results : The final data set had 29 numeric factors, with 4758 run instances from 80 participants. Participants completed between 6 and 148 (mean: 55.5) runs between November 2, 2022, and February 5, 2023. The 4 best-performing ML models were CatBoost, GMB, Light Gradient Boosting Machine, and GXBoost (average model performance > 0.90). CatBoost was the best-performing model, 96% correct in classifying runs from the injured groups. Critical power was the most important factor across all 4 algorithms, with higher values increasing the probability of classifying a run as belonging to the injured group. Conclusion : Future cohort studies using ML to explain running-related injury development should use boosting ML models, specifically CatBoost, and include critical power. We advise against interpreting these results as a causal explanation for injury development, and model performance may be overestimated due to our validation approach.

Valid surrogates are needed to monitor tibial forces during tactical load carriage training. We assessed (a) effects of load carriage on internal tibial forces, average vertical loading rate (AVLR), and peak tibial accelerations; and (b) the relationships among these variables. Walking (1.35 m/s: 0-kg, 20.4-kg, and 34.0-kg) and running (2.7 m/s: 0-kg and 20.4-kg) biomechanics were sampled (n = 34). Peak tibial force increased with load carriage during walking and running (all p < 0.001; F:25.79-371.54). AVLR increased during walking with load carriage (p < 0.001; F = 131.83), with no difference (p = 0.61; F = 0.27) when running with load carriage. Peak tibial accelerations during walking increased with 20.4-kg of load carriage (p < 0.001-0.02). When load carriage was increased to 34.0-kg, small increases were observed for peak positive acceleration (+0.1 g, p = 0.02) only, with no change in peak resultant acceleration (p = 0.99). During running, peak tibial accelerometry surprisingly decreased (p = 0.01-0.04; F = 4.69-7.35) with load carriage. Peak tibial resultant accelerometry was only weakly to moderately associated with peak tibial force during running (r = 0.38-0.52; p = 0.002-0.04); with no other associations between peak tibial force, AVLR, and peak positive tibial acceleration (r = -0.23 - 0.13; p = 0.20-0.92). Thus, AVLR and tibial acceleration are questionable surrogates for internal tibial forces during walking or running, with or without load carriage.

Load carriage training is universal during military training, regardless of sex or physical characteristics, and may contribute to the 1.3‐2.2× higher incidence of patellofemoral pain (PFP) in female versus male recruits. This study aimed to assess sex differences in patellofemoral joint (PFJ) stress during load carriage, controlling for anthropometrics and quadriceps strength. Twenty males and 20 females walked (1.35 m/s) on an instrumented treadmill with 0‐kg, 20.4‐kg, and 34.0‐kg of load carriage. An inverse‐dynamics musculoskeletal model estimated peak, impulse, and cumulative PFJ stress. To assess quadriceps strength, peak isometric knee extensor torque normalized to body mass was measured via dynamometry. Analyses of covariance (ANCOVA) adjusting for body mass, height, and quadriceps strength assessed the effects of load (0‐kg, 20.4‐kg, and 34.0‐kg) and biological sex (male, female) on PFJ stress and gait parameters. Females were shorter, had lower mass, and lower quadriceps strength (all p < 0.001; d = 1.50–1.54, indicating large effect sizes). Peak, impulse per step, and cumulative PFJ stress increased with load carriage, with greater increases in females compared to males (sex × load interactions p = 0.002–0.005; η_p² = 0.12–0.13, indicating moderate effect sizes) after controlling for body mass, height, and quadriceps strength. These data indicate that anthropometrics and quadriceps strength do not explain the substantially greater increases in per step and cumulative PFJ stress in females versus males with load carriage. Female recruits may benefit from targeted prevention efforts, such as slower progressions of load carriage training, either in amount carried or distance trained, to reduce their risk of PFP.

This systematic review and meta-analysis aims to compare physiological, perceptual and biomechanical outcomes between walking on a treadmill and overground surfaces. Five databases (CINAHL, EMBASE, MEDLINE, SPORTDiscus, Web of Science) were searched until September 2022. Included studies needed to be a crossover design comparing biomechanical, physiological, or perceptual measures between motorised-treadmill and overground walking in healthy adults (18-65 years) walking at the same speed (<5% difference). The quality of studies were assessed using a modified Downs and Black Quality Index. Meta-analyses were performed to determine standardised mean difference ± 95% confidence intervals for all main outcome measures. Fifty-five studies were included with 1,005 participants. Relative oxygen consumption (standardised mean difference [95% confidence interval] 0.38 [0.14,0.63]) and cadence (0.22 [0.06,0.38]) are higher during treadmill walking. Whereas stride length (-0.36 [-0.62,-0.11]) and step length (-0.52 [-0.98,-0.06]) are lower during treadmill walking. Most kinetic variables are different between surfaces. The oxygen consumption, spatiotemporal and kinetic differences on the treadmill may be an attempt to increase stability due to the lack of control, discomfort and familiarity on the treadmill. Treadmill construction including surface stiffness and motor power are likely additional constraints that need to be considered and require investigation. This research was supported by an Australian Government Research Training Program (RTP) scholarship. Protocol registration is CRD42020208002 (PROSPERO International Prospective Register of Systematic Reviews) in October 2020.