Garrett S. Bullock, Patrick Ward, Gary S. Collins, Tom Hughes, Franco Impellizzeri
{"title":"评论通过机器学习了解和预测足球运动中的受伤情况","authors":"Garrett S. Bullock, Patrick Ward, Gary S. Collins, Tom Hughes, Franco Impellizzeri","doi":"10.1186/s40798-024-00745-1","DOIUrl":null,"url":null,"abstract":"<p>Dear Editor,</p><p>We recently read the article titled “Machine Learning for Understanding and Predicting Injuries in Football” in Sports Medicine – Open [1]. Given that injury prediction is an emerging topic within sport, the increasing interest and excitement towards complex machine learning algorithms within this space is a cause for concern when fundamental principles of prediction model development are not followed. As such, we feel the need to intervene and highlight methodological and conceptual inaccuracies.</p><p>The models presented in this paper were deemed by the authors to be “quite sound” [1]. However, this is not true, as recently highlighted in the systematic review in Sports Medicine [2]. All of these models were included in this systematic review, and after evaluation with the established Prediction Model Risk of Bias Assessment Tool (PROBAST) [3], were rated as high or unclear risk of bias [2].</p><p>The authors detail that, “the use of machine learning has great potential to unearth new insights into the workload and injury relationship.”[1] Prediction models may use both causal and non-causal predictors to estimate the risk of a future outcome [4, 5]. Consequently, it is inappropriate to use the included predictors to infer causal relationships between individual predictors and the outcome [6, 7]. Further, the authors state that Shapley values, local interpretable model-agnostic explanations, and partial dependency plots can be used to assist in interpreting cause-effect relationships with machine learning models [8]. These tools assess for associations between predictors and outcomes and, regardless of how these methods are labelled, the popular adage “correlation is not causation” still holds [8]. Importantly, these methods remain explorative, provide post hoc explanations (rationalization), and require confirmatory studies.</p><p>Such inaccurate and incorrect interpretations of clinical prediction models are of particular concern. This is because they can lead practitioners to attempt to change injury risk by intervening or manipulating predictor variables under the false assumption of a causal relationship; while these strategies are likely ineffective, they also have potentially harmful consequences for the athlete [4, 5, 9].</p><p>While the authors promote balancing dataset outcomes through over and under-sampling [1], this is highly discouraged as ‘balancing’ datasets alters the outcome prevalence, biasing towards overestimating risk.[10, 11] Balancing data without appropriate recalibration can inappropriately impact risk prediction and ultimately decision-making [11]. The authors also encourage creating classification models. Classification models are not recommended as this supersedes clinical and performance decision-making from the model users [11]. Classification models do not allow situational context and assume all situations and individuals have the same risk threshold. Prediction models should be developed and reported as a probability or at least risk score, to allow user interpretation and decisions [11].</p><p>The authors report that area under the curve (a form of discrimination), accuracy, sensitivity, and specificity should be used to evaluate machine learning models. These metrics are only a few of the recommended performance values that should be transparently reported [12] and some (such as accuracy) have well-known problems (also in machine learning) [13, 14]. The authors do not mention calibration as a model performance metric, which is the agreement between predicted and actual probabilities [15]. Calibration is always recommended and allows users to evaluate model performance across a range of risks [4, 5, 12]. Calibration divergence at specific risk ranges can alter clinical and performance decisions, even in models with high area under the curve performance [15].</p><p>The authors mention limitations of the included research, often only evaluating a single season of injuries for a given team, and how future research should seek to evaluate the models on subsequent seasons of data. Such a statement calls into question the notion of internal and external validation of these included papers. While the models in these papers have only been internally validated, their results provide a high level of optimism but no real understanding of how they may function under different circumstances or timepoints [16, 17]. Without a process of external validation for such models it is impossible to know whether or not they will be useful to practitioners with different clubs, observing different training constraints, under different coaches, and with different athletes who may have different base levels of risk [6, 18].</p><p>The purpose of this letter is to provide clarity about an important and burgeoning sports medicine and performance field. Other points that are not examined due to word count include sample size calculations, handling of missing data, data leakage, internal validation over train test splits, and clinical utility through clinical decision analysis and impact studies. Using the highest quality methods with transparency is imperative to protect and improve athlete health and performance, and to reduce the proliferation of conceptual and methodological inaccuracies about prediction model development and interpretation.</p><p>Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.</p><ol data-track-component=\"outbound reference\" data-track-context=\"references section\"><li data-counter=\"1.\"><p>Majumdar A, Bakirov R, Hodges D, Scott S, Rees T. Machine learning for understanding and predicting injuries in football. Sports Med Open. 2022;8:73.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\"2.\"><p>Bullock GS, Mylott J, Hughes T, Nicholson KF, Riley RD, Collins GS. Just how confident can we be in predicting sports injuries? A systematic review of the methodological conduct and performance of existing musculoskeletal injury prediction models in sport. Sport Med. 2022;52(10):2469–82.</p><p>Article Google Scholar </p></li><li data-counter=\"3.\"><p>Wolff RF, Moons KG, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Int Med. 2019;170(1):51–8.</p><p>Article PubMed Google Scholar </p></li><li data-counter=\"4.\"><p>Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley R, Collins G. Methods matter: clinical prediction models will benefit sports medicine practice, but only if they are properly developed and validated. British J Sports Med. 2021;55(23):1319–21.</p><p>Article Google Scholar </p></li><li data-counter=\"5.\"><p>Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley RD, Collins GS. Clinical prediction models in sports medicine: a guide for clinicians and researchers. J Orthop Sport Phys Ther. 2021;51(10):517–25.</p><p>Article Google Scholar </p></li><li data-counter=\"6.\"><p>Bullock GS, Hughes T, Arundale AH, Ward P, Collins GS, Kluzek S. Black box prediction methods in sports medicine deserve a red card for reckless practice: a change of tactics is needed to advance athlete care. Sports Med. 2022;52(8):1729–35.</p><p>Article PubMed Google Scholar </p></li><li data-counter=\"7.\"><p>Riley RD, Hayden JA, Steyerberg EW, Moons KG, Abrams K, Kyzas PA, et al. Prognosis Research Strategy (PROGRESS) 2: prognostic factor research. PLoS Med. 2013;10(2): e1001380.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\"8.\"><p>Heskes T, Sijben E, Bucur IG, Claassen T. Causal shapley values: exploiting causal knowledge to explain individual predictions of complex models. Adv Neural Inf Process Syst. 2020;33:4778–89.</p><p>Google Scholar </p></li><li data-counter=\"9.\"><p>Impellizzeri FM, Tenan MS, Kempton T, Novak A, Coutts AJ. Acute: chronic workload ratio: conceptual issues and fundamental pitfalls. Int J Sport Phys Perform. 2020;15(6):907–13.</p><p>Article Google Scholar </p></li><li data-counter=\"10.\"><p>Harrell FE. Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis: Springer; 2001.</p></li><li data-counter=\"11.\"><p>van den Goorbergh R, van Smeden M, Timmerman D, Van Calster B. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression. J Am Med Inform Assoc. 2022;29(9):1525–34.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\"12.\"><p>Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for some traditional and novel measures. Epidemiol. 2010;21(1):128.</p><p>Article Google Scholar </p></li><li data-counter=\"13\"><p>Dinga R, Penninx BW, Veltman DJ, Schmaal L, Marquand AF. Beyond accuracy: measures for assessing machine learning models, pitfalls and guidelines. BioRxiv. 2019;8:743138.</p><p>Google Scholar </p></li><li data-counter=\"14.\"><p>Collins GS, Dhiman P, Navarro CL, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, van Smeden M. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11(7):e048008.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\"15.\"><p>Van Calster B, McLernon DJ, Van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):1–7.</p><p>Google Scholar </p></li><li data-counter=\"16\"><p>Collins GS, Altman DG. An independent external validation and evaluation of QRISK cardiovascular risk prediction: a prospective open cohort study. BMJ. 2009;7(339):b2584.</p><p>Article Google Scholar </p></li><li data-counter=\"17.\"><p>Collins GS, de Groot JA, Dutton S, Omar O, Shanyinde M, Tajar A, et al. External validation of multivariable prediction models: a systematic review of methodological conduct and reporting. BMC Med Res Methedol. 2014;14(1):1–11.</p><p>Google Scholar </p></li><li data-counter=\"18\"><p>Bullock GS, Ward P, Impellizzeri FM, Kluzek S, Hughes T, Dhiman P, et al. The trade secret taboo: open science methods are required to improve prediction models in sports medicine and performance. Sport Med. 2023;53:1–9.</p><p>Article Google Scholar </p></li></ol><p>Download references<svg aria-hidden=\"true\" focusable=\"false\" height=\"16\" role=\"img\" width=\"16\"><use xlink:href=\"#icon-eds-i-download-medium\" xmlns:xlink=\"http://www.w3.org/1999/xlink\"></use></svg></p><p>None.</p><p>No funding was obtained for this letter.</p><h3>Authors and Affiliations</h3><ol><li><p>Department of Orthopaedic Surgery and Rehabilitation, Wake Forest University School of Medicine, Winston‑Salem, NC, USA</p><p>Garrett S. Bullock</p></li><li><p>Department of Biostatistics and Data Science, Wake Forest School of Medicine, Winston-Salem, NC, USA</p><p>Garrett S. Bullock</p></li><li><p>Centre for Sport, Exercise and Osteoarthritis Research Versus Arthritis, University of Oxford, Oxford, UK</p><p>Garrett S. Bullock</p></li><li><p>Seattle Seahawks, Seattle, WA, USA</p><p>Patrick Ward</p></li><li><p>Centre for Statistics in Medicine, University of Oxford, Oxford, UK</p><p>Gary S. Collins</p></li><li><p>Department of Health Professions Institute of Sport, Manchester Metropolitan University, Manchester, UK</p><p>Tom Hughes</p></li><li><p>Institute of Sport, Manchester Metropolitan University, Manchester, UK</p><p>Tom Hughes</p></li><li><p>School of Sport, Exercise, and Rehabilitation, University of Technology Sydney, Sydney, Australia</p><p>Franco Impellizzeri</p></li></ol><span>Authors</span><ol><li><span>Garrett S. Bullock</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Patrick Ward</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Gary S. Collins</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Tom Hughes</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Franco Impellizzeri</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li></ol><h3>Contributions</h3><p>GB, PW, and FM conceived the manuscript idea. GB, PW, and FM wrote the first draft of the manuscript. GB, PW, GC, TH, and FM critically revised the manuscript. GB, PW, GC, TH, and FM approved the final version of the manuscript.</p><h3>Corresponding author</h3><p>Correspondence to Garrett S. Bullock.</p><h3>Ethics Approval and Consent to Participate</h3>\n<p>No ethics or consent approval were necessary for this letter.</p>\n<h3>Consent for Publication</h3>\n<p>Not applicable.</p>\n<h3>Competing interests</h3>\n<p>The authors declare that they have no competing interests with the content of this letter.</p><h3>Publisher's Note</h3><p>Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.</p><p><b>Open Access</b> This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.</p>\n<p>Reprints and permissions</p><img alt=\"Check for updates. Verify currency and authenticity via CrossMark\" height=\"81\" loading=\"lazy\" src=\"data:image/svg+xml;base64,<svg height="81" width="57" xmlns="http://www.w3.org/2000/svg"><g fill="none" fill-rule="evenodd"><path d="m17.35 35.45 21.3-14.2v-17.03h-21.3" fill="#989898"/><path d="m38.65 35.45-21.3-14.2v-17.03h21.3" fill="#747474"/><path d="m28 .5c-12.98 0-23.5 10.52-23.5 23.5s10.52 23.5 23.5 23.5 23.5-10.52 23.5-23.5c0-6.23-2.48-12.21-6.88-16.62-4.41-4.4-10.39-6.88-16.62-6.88zm0 41.25c-9.8 0-17.75-7.95-17.75-17.75s7.95-17.75 17.75-17.75 17.75 7.95 17.75 17.75c0 4.71-1.87 9.22-5.2 12.55s-7.84 5.2-12.55 5.2z" fill="#535353"/><path d="m41 36c-5.81 6.23-15.23 7.45-22.43 2.9-7.21-4.55-10.16-13.57-7.03-21.5l-4.92-3.11c-4.95 10.7-1.19 23.42 8.78 29.71 9.97 6.3 23.07 4.22 30.6-4.86z" fill="#9c9c9c"/><path d="m.2 58.45c0-.75.11-1.42.33-2.01s.52-1.09.91-1.5c.38-.41.83-.73 1.34-.94.51-.22 1.06-.32 1.65-.32.56 0 1.06.11 1.51.35.44.23.81.5 1.1.81l-.91 1.01c-.24-.24-.49-.42-.75-.56-.27-.13-.58-.2-.93-.2-.39 0-.73.08-1.05.23-.31.16-.58.37-.81.66-.23.28-.41.63-.53 1.04-.13.41-.19.88-.19 1.39 0 1.04.23 1.86.68 2.46.45.59 1.06.88 1.84.88.41 0 .77-.07 1.07-.23s.59-.39.85-.68l.91 1c-.38.43-.8.76-1.28.99-.47.22-1 .34-1.58.34-.59 0-1.13-.1-1.64-.31-.5-.2-.94-.51-1.31-.91-.38-.4-.67-.9-.88-1.48-.22-.59-.33-1.26-.33-2.02zm8.4-5.33h1.61v2.54l-.05 1.33c.29-.27.61-.51.96-.72s.76-.31 1.24-.31c.73 0 1.27.23 1.61.71.33.47.5 1.14.5 2.02v4.31h-1.61v-4.1c0-.57-.08-.97-.25-1.21-.17-.23-.45-.35-.83-.35-.3 0-.56.08-.79.22-.23.15-.49.36-.78.64v4.8h-1.61zm7.37 6.45c0-.56.09-1.06.26-1.51.18-.45.42-.83.71-1.14.29-.3.63-.54 1.01-.71.39-.17.78-.25 1.18-.25.47 0 .88.08 1.23.24.36.16.65.38.89.67s.42.63.54 1.03c.12.41.18.84.18 1.32 0 .32-.02.57-.07.76h-4.36c.07.62.29 1.1.65 1.44.36.33.82.5 1.38.5.29 0 .57-.04.83-.13s.51-.21.76-.37l.55 1.01c-.33.21-.69.39-1.09.53-.41.14-.83.21-1.26.21-.48 0-.92-.08-1.34-.25-.41-.16-.76-.4-1.07-.7-.31-.31-.55-.69-.72-1.13-.18-.44-.26-.95-.26-1.52zm4.6-.62c0-.55-.11-.98-.34-1.28-.23-.31-.58-.47-1.06-.47-.41 0-.77.15-1.07.45-.31.29-.5.73-.58 1.3zm2.5.62c0-.57.09-1.08.28-1.53.18-.44.43-.82.75-1.13s.69-.54 1.1-.71c.42-.16.85-.24 1.31-.24.45 0 .84.08 1.17.23s.61.34.85.57l-.77 1.02c-.19-.16-.38-.28-.56-.37-.19-.09-.39-.14-.61-.14-.56 0-1.01.21-1.35.63-.35.41-.52.97-.52 1.67 0 .69.17 1.24.51 1.66.34.41.78.62 1.32.62.28 0 .54-.06.78-.17.24-.12.45-.26.64-.42l.67 1.03c-.33.29-.69.51-1.08.65-.39.15-.78.23-1.18.23-.46 0-.9-.08-1.31-.24-.4-.16-.75-.39-1.05-.7s-.53-.69-.7-1.13c-.17-.45-.25-.96-.25-1.53zm6.91-6.45h1.58v6.17h.05l2.54-3.16h1.77l-2.35 2.8 2.59 4.07h-1.75l-1.77-2.98-1.08 1.23v1.75h-1.58zm13.69 1.27c-.25-.11-.5-.17-.75-.17-.58 0-.87.39-.87 1.16v.75h1.34v1.27h-1.34v5.6h-1.61v-5.6h-.92v-1.2l.92-.07v-.72c0-.35.04-.68.13-.98.08-.31.21-.57.4-.79s.42-.39.71-.51c.28-.12.63-.18 1.04-.18.24 0 .48.02.69.07.22.05.41.1.57.17zm.48 5.18c0-.57.09-1.08.27-1.53.17-.44.41-.82.72-1.13.3-.31.65-.54 1.04-.71.39-.16.8-.24 1.23-.24s.84.08 1.24.24c.4.17.74.4 1.04.71s.54.69.72 1.13c.19.45.28.96.28 1.53s-.09 1.08-.28 1.53c-.18.44-.42.82-.72 1.13s-.64.54-1.04.7-.81.24-1.24.24-.84-.08-1.23-.24-.74-.39-1.04-.7c-.31-.31-.55-.69-.72-1.13-.18-.45-.27-.96-.27-1.53zm1.65 0c0 .69.14 1.24.43 1.66.28.41.68.62 1.18.62.51 0 .9-.21 1.19-.62.29-.42.44-.97.44-1.66 0-.7-.15-1.26-.44-1.67-.29-.42-.68-.63-1.19-.63-.5 0-.9.21-1.18.63-.29.41-.43.97-.43 1.67zm6.48-3.44h1.33l.12 1.21h.05c.24-.44.54-.79.88-1.02.35-.24.7-.36 1.07-.36.32 0 .59.05.78.14l-.28 1.4-.33-.09c-.11-.01-.23-.02-.38-.02-.27 0-.56.1-.86.31s-.55.58-.77 1.1v4.2h-1.61zm-47.87 15h1.61v4.1c0 .57.08.97.25 1.2.17.24.44.35.81.35.3 0 .57-.07.8-.22.22-.15.47-.39.73-.73v-4.7h1.61v6.87h-1.32l-.12-1.01h-.04c-.3.36-.63.64-.98.86-.35.21-.76.32-1.24.32-.73 0-1.27-.24-1.61-.71-.33-.47-.5-1.14-.5-2.02zm9.46 7.43v2.16h-1.61v-9.59h1.33l.12.72h.05c.29-.24.61-.45.97-.63.35-.17.72-.26 1.1-.26.43 0 .81.08 1.15.24.33.17.61.4.84.71.24.31.41.68.53 1.11.13.42.19.91.19 1.44 0 .59-.09 1.11-.25 1.57-.16.47-.38.85-.65 1.16-.27.32-.58.56-.94.73-.35.16-.72.25-1.1.25-.3 0-.6-.07-.9-.2s-.59-.31-.87-.56zm0-2.3c.26.22.5.37.73.45.24.09.46.13.66.13.46 0 .84-.2 1.15-.6.31-.39.46-.98.46-1.77 0-.69-.12-1.22-.35-1.61-.23-.38-.61-.57-1.13-.57-.49 0-.99.26-1.52.77zm5.87-1.69c0-.56.08-1.06.25-1.51.16-.45.37-.83.65-1.14.27-.3.58-.54.93-.71s.71-.25 1.08-.25c.39 0 .73.07 1 .2.27.14.54.32.81.55l-.06-1.1v-2.49h1.61v9.88h-1.33l-.11-.74h-.06c-.25.25-.54.46-.88.64-.33.18-.69.27-1.06.27-.87 0-1.56-.32-2.07-.95s-.76-1.51-.76-2.65zm1.67-.01c0 .74.13 1.31.4 1.7.26.38.65.58 1.15.58.51 0 .99-.26 1.44-.77v-3.21c-.24-.21-.48-.36-.7-.45-.23-.08-.46-.12-.7-.12-.45 0-.82.19-1.13.59-.31.39-.46.95-.46 1.68zm6.35 1.59c0-.73.32-1.3.97-1.71.64-.4 1.67-.68 3.08-.84 0-.17-.02-.34-.07-.51-.05-.16-.12-.3-.22-.43s-.22-.22-.38-.3c-.15-.06-.34-.1-.58-.1-.34 0-.68.07-1 .2s-.63.29-.93.47l-.59-1.08c.39-.24.81-.45 1.28-.63.47-.17.99-.26 1.54-.26.86 0 1.51.25 1.93.76s.63 1.25.63 2.21v4.07h-1.32l-.12-.76h-.05c-.3.27-.63.48-.98.66s-.73.27-1.14.27c-.61 0-1.1-.19-1.48-.56-.38-.36-.57-.85-.57-1.46zm1.57-.12c0 .3.09.53.27.67.19.14.42.21.71.21.28 0 .54-.07.77-.2s.48-.31.73-.56v-1.54c-.47.06-.86.13-1.18.23-.31.09-.57.19-.76.31s-.33.25-.41.4c-.09.15-.13.31-.13.48zm6.29-3.63h-.98v-1.2l1.06-.07.2-1.88h1.34v1.88h1.75v1.27h-1.75v3.28c0 .8.32 1.2.97 1.2.12 0 .24-.01.37-.04.12-.03.24-.07.34-.11l.28 1.19c-.19.06-.4.12-.64.17-.23.05-.49.08-.76.08-.4 0-.74-.06-1.02-.18-.27-.13-.49-.3-.67-.52-.17-.21-.3-.48-.37-.78-.08-.3-.12-.64-.12-1.01zm4.36 2.17c0-.56.09-1.06.27-1.51s.41-.83.71-1.14c.29-.3.63-.54 1.01-.71.39-.17.78-.25 1.18-.25.47 0 .88.08 1.23.24.36.16.65.38.89.67s.42.63.54 1.03c.12.41.18.84.18 1.32 0 .32-.02.57-.07.76h-4.37c.08.62.29 1.1.65 1.44.36.33.82.5 1.38.5.3 0 .58-.04.84-.13.25-.09.51-.21.76-.37l.54 1.01c-.32.21-.69.39-1.09.53s-.82.21-1.26.21c-.47 0-.92-.08-1.33-.25-.41-.16-.77-.4-1.08-.7-.3-.31-.54-.69-.72-1.13-.17-.44-.26-.95-.26-1.52zm4.61-.62c0-.55-.11-.98-.34-1.28-.23-.31-.58-.47-1.06-.47-.41 0-.77.15-1.08.45-.31.29-.5.73-.57 1.3zm3.01 2.23c.31.24.61.43.92.57.3.13.63.2.98.2.38 0 .65-.08.83-.23s.27-.35.27-.6c0-.14-.05-.26-.13-.37-.08-.1-.2-.2-.34-.28-.14-.09-.29-.16-.47-.23l-.53-.22c-.23-.09-.46-.18-.69-.3-.23-.11-.44-.24-.62-.4s-.33-.35-.45-.55c-.12-.21-.18-.46-.18-.75 0-.61.23-1.1.68-1.49.44-.38 1.06-.57 1.83-.57.48 0 .91.08 1.29.25s.71.36.99.57l-.74.98c-.24-.17-.49-.32-.73-.42-.25-.11-.51-.16-.78-.16-.35 0-.6.07-.76.21-.17.15-.25.33-.25.54 0 .14.04.26.12.36s.18.18.31.26c.14.07.29.14.46.21l.54.19c.23.09.47.18.7.29s.44.24.64.4c.19.16.34.35.46.58.11.23.17.5.17.82 0 .3-.06.58-.17.83-.12.26-.29.48-.51.68-.23.19-.51.34-.84.45-.34.11-.72.17-1.15.17-.48 0-.95-.09-1.41-.27-.46-.19-.86-.41-1.2-.68z" fill="#535353"/></g></svg>\" width=\"57\"/><h3>Cite this article</h3><p>Bullock, G.S., Ward, P., Collins, G.S. <i>et al.</i> Comment on: Machine Learning for Understanding and Predicting Injuries in Football. <i>Sports Med - Open</i> <b>10</b>, 84 (2024). https://doi.org/10.1186/s40798-024-00745-1</p><p>Download citation<svg aria-hidden=\"true\" focusable=\"false\" height=\"16\" role=\"img\" width=\"16\"><use xlink:href=\"#icon-eds-i-download-medium\" xmlns:xlink=\"http://www.w3.org/1999/xlink\"></use></svg></p><ul data-test=\"publication-history\"><li><p>Received<span>: </span><span><time datetime=\"2024-01-23\">23 January 2024</time></span></p></li><li><p>Accepted<span>: </span><span><time datetime=\"2024-06-25\">25 June 2024</time></span></p></li><li><p>Published<span>: </span><span><time datetime=\"2024-07-28\">28 July 2024</time></span></p></li><li><p>DOI</abbr><span>: </span><span>https://doi.org/10.1186/s40798-024-00745-1</span></p></li></ul><h3>Share this article</h3><p>Anyone you share the following link with will be able to read this content:</p><button data-track=\"click\" data-track-action=\"get shareable link\" data-track-external=\"\" data-track-label=\"button\" type=\"button\">Get shareable link</button><p>Sorry, a shareable link is not currently available for this article.</p><p data-track=\"click\" data-track-action=\"select share url\" data-track-label=\"button\"></p><button data-track=\"click\" data-track-action=\"copy share url\" data-track-external=\"\" data-track-label=\"button\" type=\"button\">Copy to clipboard</button><p> Provided by the Springer Nature SharedIt content-sharing initiative </p>","PeriodicalId":21788,"journal":{"name":"Sports Medicine - Open","volume":"69 1","pages":""},"PeriodicalIF":4.1000,"publicationDate":"2024-07-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Comment on: Machine Learning for Understanding and Predicting Injuries in Football\",\"authors\":\"Garrett S. Bullock, Patrick Ward, Gary S. Collins, Tom Hughes, Franco Impellizzeri\",\"doi\":\"10.1186/s40798-024-00745-1\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>Dear Editor,</p><p>We recently read the article titled “Machine Learning for Understanding and Predicting Injuries in Football” in Sports Medicine – Open [1]. Given that injury prediction is an emerging topic within sport, the increasing interest and excitement towards complex machine learning algorithms within this space is a cause for concern when fundamental principles of prediction model development are not followed. As such, we feel the need to intervene and highlight methodological and conceptual inaccuracies.</p><p>The models presented in this paper were deemed by the authors to be “quite sound” [1]. However, this is not true, as recently highlighted in the systematic review in Sports Medicine [2]. All of these models were included in this systematic review, and after evaluation with the established Prediction Model Risk of Bias Assessment Tool (PROBAST) [3], were rated as high or unclear risk of bias [2].</p><p>The authors detail that, “the use of machine learning has great potential to unearth new insights into the workload and injury relationship.”[1] Prediction models may use both causal and non-causal predictors to estimate the risk of a future outcome [4, 5]. Consequently, it is inappropriate to use the included predictors to infer causal relationships between individual predictors and the outcome [6, 7]. Further, the authors state that Shapley values, local interpretable model-agnostic explanations, and partial dependency plots can be used to assist in interpreting cause-effect relationships with machine learning models [8]. These tools assess for associations between predictors and outcomes and, regardless of how these methods are labelled, the popular adage “correlation is not causation” still holds [8]. Importantly, these methods remain explorative, provide post hoc explanations (rationalization), and require confirmatory studies.</p><p>Such inaccurate and incorrect interpretations of clinical prediction models are of particular concern. This is because they can lead practitioners to attempt to change injury risk by intervening or manipulating predictor variables under the false assumption of a causal relationship; while these strategies are likely ineffective, they also have potentially harmful consequences for the athlete [4, 5, 9].</p><p>While the authors promote balancing dataset outcomes through over and under-sampling [1], this is highly discouraged as ‘balancing’ datasets alters the outcome prevalence, biasing towards overestimating risk.[10, 11] Balancing data without appropriate recalibration can inappropriately impact risk prediction and ultimately decision-making [11]. The authors also encourage creating classification models. Classification models are not recommended as this supersedes clinical and performance decision-making from the model users [11]. Classification models do not allow situational context and assume all situations and individuals have the same risk threshold. Prediction models should be developed and reported as a probability or at least risk score, to allow user interpretation and decisions [11].</p><p>The authors report that area under the curve (a form of discrimination), accuracy, sensitivity, and specificity should be used to evaluate machine learning models. These metrics are only a few of the recommended performance values that should be transparently reported [12] and some (such as accuracy) have well-known problems (also in machine learning) [13, 14]. The authors do not mention calibration as a model performance metric, which is the agreement between predicted and actual probabilities [15]. Calibration is always recommended and allows users to evaluate model performance across a range of risks [4, 5, 12]. Calibration divergence at specific risk ranges can alter clinical and performance decisions, even in models with high area under the curve performance [15].</p><p>The authors mention limitations of the included research, often only evaluating a single season of injuries for a given team, and how future research should seek to evaluate the models on subsequent seasons of data. Such a statement calls into question the notion of internal and external validation of these included papers. While the models in these papers have only been internally validated, their results provide a high level of optimism but no real understanding of how they may function under different circumstances or timepoints [16, 17]. Without a process of external validation for such models it is impossible to know whether or not they will be useful to practitioners with different clubs, observing different training constraints, under different coaches, and with different athletes who may have different base levels of risk [6, 18].</p><p>The purpose of this letter is to provide clarity about an important and burgeoning sports medicine and performance field. Other points that are not examined due to word count include sample size calculations, handling of missing data, data leakage, internal validation over train test splits, and clinical utility through clinical decision analysis and impact studies. Using the highest quality methods with transparency is imperative to protect and improve athlete health and performance, and to reduce the proliferation of conceptual and methodological inaccuracies about prediction model development and interpretation.</p><p>Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.</p><ol data-track-component=\\\"outbound reference\\\" data-track-context=\\\"references section\\\"><li data-counter=\\\"1.\\\"><p>Majumdar A, Bakirov R, Hodges D, Scott S, Rees T. Machine learning for understanding and predicting injuries in football. Sports Med Open. 2022;8:73.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\\\"2.\\\"><p>Bullock GS, Mylott J, Hughes T, Nicholson KF, Riley RD, Collins GS. Just how confident can we be in predicting sports injuries? A systematic review of the methodological conduct and performance of existing musculoskeletal injury prediction models in sport. Sport Med. 2022;52(10):2469–82.</p><p>Article Google Scholar </p></li><li data-counter=\\\"3.\\\"><p>Wolff RF, Moons KG, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Int Med. 2019;170(1):51–8.</p><p>Article PubMed Google Scholar </p></li><li data-counter=\\\"4.\\\"><p>Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley R, Collins G. Methods matter: clinical prediction models will benefit sports medicine practice, but only if they are properly developed and validated. British J Sports Med. 2021;55(23):1319–21.</p><p>Article Google Scholar </p></li><li data-counter=\\\"5.\\\"><p>Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley RD, Collins GS. Clinical prediction models in sports medicine: a guide for clinicians and researchers. J Orthop Sport Phys Ther. 2021;51(10):517–25.</p><p>Article Google Scholar </p></li><li data-counter=\\\"6.\\\"><p>Bullock GS, Hughes T, Arundale AH, Ward P, Collins GS, Kluzek S. Black box prediction methods in sports medicine deserve a red card for reckless practice: a change of tactics is needed to advance athlete care. Sports Med. 2022;52(8):1729–35.</p><p>Article PubMed Google Scholar </p></li><li data-counter=\\\"7.\\\"><p>Riley RD, Hayden JA, Steyerberg EW, Moons KG, Abrams K, Kyzas PA, et al. Prognosis Research Strategy (PROGRESS) 2: prognostic factor research. PLoS Med. 2013;10(2): e1001380.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\\\"8.\\\"><p>Heskes T, Sijben E, Bucur IG, Claassen T. Causal shapley values: exploiting causal knowledge to explain individual predictions of complex models. Adv Neural Inf Process Syst. 2020;33:4778–89.</p><p>Google Scholar </p></li><li data-counter=\\\"9.\\\"><p>Impellizzeri FM, Tenan MS, Kempton T, Novak A, Coutts AJ. Acute: chronic workload ratio: conceptual issues and fundamental pitfalls. Int J Sport Phys Perform. 2020;15(6):907–13.</p><p>Article Google Scholar </p></li><li data-counter=\\\"10.\\\"><p>Harrell FE. Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis: Springer; 2001.</p></li><li data-counter=\\\"11.\\\"><p>van den Goorbergh R, van Smeden M, Timmerman D, Van Calster B. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression. J Am Med Inform Assoc. 2022;29(9):1525–34.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\\\"12.\\\"><p>Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for some traditional and novel measures. Epidemiol. 2010;21(1):128.</p><p>Article Google Scholar </p></li><li data-counter=\\\"13\\\"><p>Dinga R, Penninx BW, Veltman DJ, Schmaal L, Marquand AF. Beyond accuracy: measures for assessing machine learning models, pitfalls and guidelines. BioRxiv. 2019;8:743138.</p><p>Google Scholar </p></li><li data-counter=\\\"14.\\\"><p>Collins GS, Dhiman P, Navarro CL, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, van Smeden M. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11(7):e048008.</p><p>Article PubMed PubMed Central Google Scholar </p></li><li data-counter=\\\"15.\\\"><p>Van Calster B, McLernon DJ, Van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):1–7.</p><p>Google Scholar </p></li><li data-counter=\\\"16\\\"><p>Collins GS, Altman DG. An independent external validation and evaluation of QRISK cardiovascular risk prediction: a prospective open cohort study. BMJ. 2009;7(339):b2584.</p><p>Article Google Scholar </p></li><li data-counter=\\\"17.\\\"><p>Collins GS, de Groot JA, Dutton S, Omar O, Shanyinde M, Tajar A, et al. External validation of multivariable prediction models: a systematic review of methodological conduct and reporting. BMC Med Res Methedol. 2014;14(1):1–11.</p><p>Google Scholar </p></li><li data-counter=\\\"18\\\"><p>Bullock GS, Ward P, Impellizzeri FM, Kluzek S, Hughes T, Dhiman P, et al. The trade secret taboo: open science methods are required to improve prediction models in sports medicine and performance. Sport Med. 2023;53:1–9.</p><p>Article Google Scholar </p></li></ol><p>Download references<svg aria-hidden=\\\"true\\\" focusable=\\\"false\\\" height=\\\"16\\\" role=\\\"img\\\" width=\\\"16\\\"><use xlink:href=\\\"#icon-eds-i-download-medium\\\" xmlns:xlink=\\\"http://www.w3.org/1999/xlink\\\"></use></svg></p><p>None.</p><p>No funding was obtained for this letter.</p><h3>Authors and Affiliations</h3><ol><li><p>Department of Orthopaedic Surgery and Rehabilitation, Wake Forest University School of Medicine, Winston‑Salem, NC, USA</p><p>Garrett S. Bullock</p></li><li><p>Department of Biostatistics and Data Science, Wake Forest School of Medicine, Winston-Salem, NC, USA</p><p>Garrett S. Bullock</p></li><li><p>Centre for Sport, Exercise and Osteoarthritis Research Versus Arthritis, University of Oxford, Oxford, UK</p><p>Garrett S. Bullock</p></li><li><p>Seattle Seahawks, Seattle, WA, USA</p><p>Patrick Ward</p></li><li><p>Centre for Statistics in Medicine, University of Oxford, Oxford, UK</p><p>Gary S. Collins</p></li><li><p>Department of Health Professions Institute of Sport, Manchester Metropolitan University, Manchester, UK</p><p>Tom Hughes</p></li><li><p>Institute of Sport, Manchester Metropolitan University, Manchester, UK</p><p>Tom Hughes</p></li><li><p>School of Sport, Exercise, and Rehabilitation, University of Technology Sydney, Sydney, Australia</p><p>Franco Impellizzeri</p></li></ol><span>Authors</span><ol><li><span>Garrett S. Bullock</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Patrick Ward</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Gary S. Collins</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Tom Hughes</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li><li><span>Franco Impellizzeri</span>View author publications<p>You can also search for this author in <span>PubMed<span> </span>Google Scholar</span></p></li></ol><h3>Contributions</h3><p>GB, PW, and FM conceived the manuscript idea. GB, PW, and FM wrote the first draft of the manuscript. GB, PW, GC, TH, and FM critically revised the manuscript. GB, PW, GC, TH, and FM approved the final version of the manuscript.</p><h3>Corresponding author</h3><p>Correspondence to Garrett S. Bullock.</p><h3>Ethics Approval and Consent to Participate</h3>\\n<p>No ethics or consent approval were necessary for this letter.</p>\\n<h3>Consent for Publication</h3>\\n<p>Not applicable.</p>\\n<h3>Competing interests</h3>\\n<p>The authors declare that they have no competing interests with the content of this letter.</p><h3>Publisher's Note</h3><p>Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.</p><p><b>Open Access</b> This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.</p>\\n<p>Reprints and permissions</p><img alt=\\\"Check for updates. Verify currency and authenticity via CrossMark\\\" height=\\\"81\\\" loading=\\\"lazy\\\" src=\\\"data:image/svg+xml;base64,<svg height="81" width="57" xmlns="http://www.w3.org/2000/svg"><g fill="none" fill-rule="evenodd"><path d="m17.35 35.45 21.3-14.2v-17.03h-21.3" fill="#989898"/><path d="m38.65 35.45-21.3-14.2v-17.03h21.3" fill="#747474"/><path d="m28 .5c-12.98 0-23.5 10.52-23.5 23.5s10.52 23.5 23.5 23.5 23.5-10.52 23.5-23.5c0-6.23-2.48-12.21-6.88-16.62-4.41-4.4-10.39-6.88-16.62-6.88zm0 41.25c-9.8 0-17.75-7.95-17.75-17.75s7.95-17.75 17.75-17.75 17.75 7.95 17.75 17.75c0 4.71-1.87 9.22-5.2 12.55s-7.84 5.2-12.55 5.2z" fill="#535353"/><path d="m41 36c-5.81 6.23-15.23 7.45-22.43 2.9-7.21-4.55-10.16-13.57-7.03-21.5l-4.92-3.11c-4.95 10.7-1.19 23.42 8.78 29.71 9.97 6.3 23.07 4.22 30.6-4.86z" fill="#9c9c9c"/><path d="m.2 58.45c0-.75.11-1.42.33-2.01s.52-1.09.91-1.5c.38-.41.83-.73 1.34-.94.51-.22 1.06-.32 1.65-.32.56 0 1.06.11 1.51.35.44.23.81.5 1.1.81l-.91 1.01c-.24-.24-.49-.42-.75-.56-.27-.13-.58-.2-.93-.2-.39 0-.73.08-1.05.23-.31.16-.58.37-.81.66-.23.28-.41.63-.53 1.04-.13.41-.19.88-.19 1.39 0 1.04.23 1.86.68 2.46.45.59 1.06.88 1.84.88.41 0 .77-.07 1.07-.23s.59-.39.85-.68l.91 1c-.38.43-.8.76-1.28.99-.47.22-1 .34-1.58.34-.59 0-1.13-.1-1.64-.31-.5-.2-.94-.51-1.31-.91-.38-.4-.67-.9-.88-1.48-.22-.59-.33-1.26-.33-2.02zm8.4-5.33h1.61v2.54l-.05 1.33c.29-.27.61-.51.96-.72s.76-.31 1.24-.31c.73 0 1.27.23 1.61.71.33.47.5 1.14.5 2.02v4.31h-1.61v-4.1c0-.57-.08-.97-.25-1.21-.17-.23-.45-.35-.83-.35-.3 0-.56.08-.79.22-.23.15-.49.36-.78.64v4.8h-1.61zm7.37 6.45c0-.56.09-1.06.26-1.51.18-.45.42-.83.71-1.14.29-.3.63-.54 1.01-.71.39-.17.78-.25 1.18-.25.47 0 .88.08 1.23.24.36.16.65.38.89.67s.42.63.54 1.03c.12.41.18.84.18 1.32 0 .32-.02.57-.07.76h-4.36c.07.62.29 1.1.65 1.44.36.33.82.5 1.38.5.29 0 .57-.04.83-.13s.51-.21.76-.37l.55 1.01c-.33.21-.69.39-1.09.53-.41.14-.83.21-1.26.21-.48 0-.92-.08-1.34-.25-.41-.16-.76-.4-1.07-.7-.31-.31-.55-.69-.72-1.13-.18-.44-.26-.95-.26-1.52zm4.6-.62c0-.55-.11-.98-.34-1.28-.23-.31-.58-.47-1.06-.47-.41 0-.77.15-1.07.45-.31.29-.5.73-.58 1.3zm2.5.62c0-.57.09-1.08.28-1.53.18-.44.43-.82.75-1.13s.69-.54 1.1-.71c.42-.16.85-.24 1.31-.24.45 0 .84.08 1.17.23s.61.34.85.57l-.77 1.02c-.19-.16-.38-.28-.56-.37-.19-.09-.39-.14-.61-.14-.56 0-1.01.21-1.35.63-.35.41-.52.97-.52 1.67 0 .69.17 1.24.51 1.66.34.41.78.62 1.32.62.28 0 .54-.06.78-.17.24-.12.45-.26.64-.42l.67 1.03c-.33.29-.69.51-1.08.65-.39.15-.78.23-1.18.23-.46 0-.9-.08-1.31-.24-.4-.16-.75-.39-1.05-.7s-.53-.69-.7-1.13c-.17-.45-.25-.96-.25-1.53zm6.91-6.45h1.58v6.17h.05l2.54-3.16h1.77l-2.35 2.8 2.59 4.07h-1.75l-1.77-2.98-1.08 1.23v1.75h-1.58zm13.69 1.27c-.25-.11-.5-.17-.75-.17-.58 0-.87.39-.87 1.16v.75h1.34v1.27h-1.34v5.6h-1.61v-5.6h-.92v-1.2l.92-.07v-.72c0-.35.04-.68.13-.98.08-.31.21-.57.4-.79s.42-.39.71-.51c.28-.12.63-.18 1.04-.18.24 0 .48.02.69.07.22.05.41.1.57.17zm.48 5.18c0-.57.09-1.08.27-1.53.17-.44.41-.82.72-1.13.3-.31.65-.54 1.04-.71.39-.16.8-.24 1.23-.24s.84.08 1.24.24c.4.17.74.4 1.04.71s.54.69.72 1.13c.19.45.28.96.28 1.53s-.09 1.08-.28 1.53c-.18.44-.42.82-.72 1.13s-.64.54-1.04.7-.81.24-1.24.24-.84-.08-1.23-.24-.74-.39-1.04-.7c-.31-.31-.55-.69-.72-1.13-.18-.45-.27-.96-.27-1.53zm1.65 0c0 .69.14 1.24.43 1.66.28.41.68.62 1.18.62.51 0 .9-.21 1.19-.62.29-.42.44-.97.44-1.66 0-.7-.15-1.26-.44-1.67-.29-.42-.68-.63-1.19-.63-.5 0-.9.21-1.18.63-.29.41-.43.97-.43 1.67zm6.48-3.44h1.33l.12 1.21h.05c.24-.44.54-.79.88-1.02.35-.24.7-.36 1.07-.36.32 0 .59.05.78.14l-.28 1.4-.33-.09c-.11-.01-.23-.02-.38-.02-.27 0-.56.1-.86.31s-.55.58-.77 1.1v4.2h-1.61zm-47.87 15h1.61v4.1c0 .57.08.97.25 1.2.17.24.44.35.81.35.3 0 .57-.07.8-.22.22-.15.47-.39.73-.73v-4.7h1.61v6.87h-1.32l-.12-1.01h-.04c-.3.36-.63.64-.98.86-.35.21-.76.32-1.24.32-.73 0-1.27-.24-1.61-.71-.33-.47-.5-1.14-.5-2.02zm9.46 7.43v2.16h-1.61v-9.59h1.33l.12.72h.05c.29-.24.61-.45.97-.63.35-.17.72-.26 1.1-.26.43 0 .81.08 1.15.24.33.17.61.4.84.71.24.31.41.68.53 1.11.13.42.19.91.19 1.44 0 .59-.09 1.11-.25 1.57-.16.47-.38.85-.65 1.16-.27.32-.58.56-.94.73-.35.16-.72.25-1.1.25-.3 0-.6-.07-.9-.2s-.59-.31-.87-.56zm0-2.3c.26.22.5.37.73.45.24.09.46.13.66.13.46 0 .84-.2 1.15-.6.31-.39.46-.98.46-1.77 0-.69-.12-1.22-.35-1.61-.23-.38-.61-.57-1.13-.57-.49 0-.99.26-1.52.77zm5.87-1.69c0-.56.08-1.06.25-1.51.16-.45.37-.83.65-1.14.27-.3.58-.54.93-.71s.71-.25 1.08-.25c.39 0 .73.07 1 .2.27.14.54.32.81.55l-.06-1.1v-2.49h1.61v9.88h-1.33l-.11-.74h-.06c-.25.25-.54.46-.88.64-.33.18-.69.27-1.06.27-.87 0-1.56-.32-2.07-.95s-.76-1.51-.76-2.65zm1.67-.01c0 .74.13 1.31.4 1.7.26.38.65.58 1.15.58.51 0 .99-.26 1.44-.77v-3.21c-.24-.21-.48-.36-.7-.45-.23-.08-.46-.12-.7-.12-.45 0-.82.19-1.13.59-.31.39-.46.95-.46 1.68zm6.35 1.59c0-.73.32-1.3.97-1.71.64-.4 1.67-.68 3.08-.84 0-.17-.02-.34-.07-.51-.05-.16-.12-.3-.22-.43s-.22-.22-.38-.3c-.15-.06-.34-.1-.58-.1-.34 0-.68.07-1 .2s-.63.29-.93.47l-.59-1.08c.39-.24.81-.45 1.28-.63.47-.17.99-.26 1.54-.26.86 0 1.51.25 1.93.76s.63 1.25.63 2.21v4.07h-1.32l-.12-.76h-.05c-.3.27-.63.48-.98.66s-.73.27-1.14.27c-.61 0-1.1-.19-1.48-.56-.38-.36-.57-.85-.57-1.46zm1.57-.12c0 .3.09.53.27.67.19.14.42.21.71.21.28 0 .54-.07.77-.2s.48-.31.73-.56v-1.54c-.47.06-.86.13-1.18.23-.31.09-.57.19-.76.31s-.33.25-.41.4c-.09.15-.13.31-.13.48zm6.29-3.63h-.98v-1.2l1.06-.07.2-1.88h1.34v1.88h1.75v1.27h-1.75v3.28c0 .8.32 1.2.97 1.2.12 0 .24-.01.37-.04.12-.03.24-.07.34-.11l.28 1.19c-.19.06-.4.12-.64.17-.23.05-.49.08-.76.08-.4 0-.74-.06-1.02-.18-.27-.13-.49-.3-.67-.52-.17-.21-.3-.48-.37-.78-.08-.3-.12-.64-.12-1.01zm4.36 2.17c0-.56.09-1.06.27-1.51s.41-.83.71-1.14c.29-.3.63-.54 1.01-.71.39-.17.78-.25 1.18-.25.47 0 .88.08 1.23.24.36.16.65.38.89.67s.42.63.54 1.03c.12.41.18.84.18 1.32 0 .32-.02.57-.07.76h-4.37c.08.62.29 1.1.65 1.44.36.33.82.5 1.38.5.3 0 .58-.04.84-.13.25-.09.51-.21.76-.37l.54 1.01c-.32.21-.69.39-1.09.53s-.82.21-1.26.21c-.47 0-.92-.08-1.33-.25-.41-.16-.77-.4-1.08-.7-.3-.31-.54-.69-.72-1.13-.17-.44-.26-.95-.26-1.52zm4.61-.62c0-.55-.11-.98-.34-1.28-.23-.31-.58-.47-1.06-.47-.41 0-.77.15-1.08.45-.31.29-.5.73-.57 1.3zm3.01 2.23c.31.24.61.43.92.57.3.13.63.2.98.2.38 0 .65-.08.83-.23s.27-.35.27-.6c0-.14-.05-.26-.13-.37-.08-.1-.2-.2-.34-.28-.14-.09-.29-.16-.47-.23l-.53-.22c-.23-.09-.46-.18-.69-.3-.23-.11-.44-.24-.62-.4s-.33-.35-.45-.55c-.12-.21-.18-.46-.18-.75 0-.61.23-1.1.68-1.49.44-.38 1.06-.57 1.83-.57.48 0 .91.08 1.29.25s.71.36.99.57l-.74.98c-.24-.17-.49-.32-.73-.42-.25-.11-.51-.16-.78-.16-.35 0-.6.07-.76.21-.17.15-.25.33-.25.54 0 .14.04.26.12.36s.18.18.31.26c.14.07.29.14.46.21l.54.19c.23.09.47.18.7.29s.44.24.64.4c.19.16.34.35.46.58.11.23.17.5.17.82 0 .3-.06.58-.17.83-.12.26-.29.48-.51.68-.23.19-.51.34-.84.45-.34.11-.72.17-1.15.17-.48 0-.95-.09-1.41-.27-.46-.19-.86-.41-1.2-.68z" fill="#535353"/></g></svg>\\\" width=\\\"57\\\"/><h3>Cite this article</h3><p>Bullock, G.S., Ward, P., Collins, G.S. <i>et al.</i> Comment on: Machine Learning for Understanding and Predicting Injuries in Football. <i>Sports Med - Open</i> <b>10</b>, 84 (2024). https://doi.org/10.1186/s40798-024-00745-1</p><p>Download citation<svg aria-hidden=\\\"true\\\" focusable=\\\"false\\\" height=\\\"16\\\" role=\\\"img\\\" width=\\\"16\\\"><use xlink:href=\\\"#icon-eds-i-download-medium\\\" xmlns:xlink=\\\"http://www.w3.org/1999/xlink\\\"></use></svg></p><ul data-test=\\\"publication-history\\\"><li><p>Received<span>: </span><span><time datetime=\\\"2024-01-23\\\">23 January 2024</time></span></p></li><li><p>Accepted<span>: </span><span><time datetime=\\\"2024-06-25\\\">25 June 2024</time></span></p></li><li><p>Published<span>: </span><span><time datetime=\\\"2024-07-28\\\">28 July 2024</time></span></p></li><li><p>DOI</abbr><span>: </span><span>https://doi.org/10.1186/s40798-024-00745-1</span></p></li></ul><h3>Share this article</h3><p>Anyone you share the following link with will be able to read this content:</p><button data-track=\\\"click\\\" data-track-action=\\\"get shareable link\\\" data-track-external=\\\"\\\" data-track-label=\\\"button\\\" type=\\\"button\\\">Get shareable link</button><p>Sorry, a shareable link is not currently available for this article.</p><p data-track=\\\"click\\\" data-track-action=\\\"select share url\\\" data-track-label=\\\"button\\\"></p><button data-track=\\\"click\\\" data-track-action=\\\"copy share url\\\" data-track-external=\\\"\\\" data-track-label=\\\"button\\\" type=\\\"button\\\">Copy to clipboard</button><p> Provided by the Springer Nature SharedIt content-sharing initiative </p>\",\"PeriodicalId\":21788,\"journal\":{\"name\":\"Sports Medicine - Open\",\"volume\":\"69 1\",\"pages\":\"\"},\"PeriodicalIF\":4.1000,\"publicationDate\":\"2024-07-28\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Sports Medicine - Open\",\"FirstCategoryId\":\"3\",\"ListUrlMain\":\"https://doi.org/10.1186/s40798-024-00745-1\",\"RegionNum\":2,\"RegionCategory\":\"医学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"SPORT SCIENCES\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Sports Medicine - Open","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1186/s40798-024-00745-1","RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"SPORT SCIENCES","Score":null,"Total":0}
引用次数: 0
摘要
由于字数原因,未对其他要点进行研究,包括样本量计算、缺失数据的处理、数据泄漏、训练测试分裂的内部验证,以及通过临床决策分析和影响研究实现的临床实用性。数据共享不适用于本文,因为本研究未生成或分析数据集。Majumdar A, Bakirov R, Hodges D, Scott S, Rees T. Machine learning for understanding and predicting injuries in football.Sports Med Open.2022;8:73.Article PubMed PubMed Central Google Scholar Bullock GS, Mylott J, Hughes T, Nicholson KF, Riley RD, Collins GS.我们对运动损伤的预测到底有多大把握?对体育运动中现有肌肉骨骼损伤预测模型的方法和性能的系统回顾。Sport Med.2022;52(10):2469-82.Article Google Scholar Wolff RF, Moons KG, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies.Ann Int Med.2019;170(1):51-8.Article PubMed Google Scholar Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley R, Collins G. Methods matter: clinical prediction models will benefit sports medicine practice, but only if they are properly developed and validated.British J Sports Med.2021;55(23):1319-21.Article Google Scholar Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley RD, Collins GS.运动医学临床预测模型:临床医生和研究人员指南》。J Orthop Sport Phys Ther.2021;51(10):517-25.Article Google Scholar Bullock GS, Hughes T, Arundale AH, Ward P, Collins GS, Kluzek S. Black box prediction methods in sports medicine deserve a red card for reckless practice: a change of tactics is needed to advance athlete care.运动医学》。2022; 52(8):1729-35.Article PubMed Google Scholar Riley RD, Hayden JA, Steyerberg EW, Moons KG, Abrams K, Kyzas PA, et al. Prognosis Research Strategy (PROGRESS) 2: prognostic factor research.PLoS Med.2013;10(2): e1001380.Article PubMed PubMed Central Google Scholar Heskes T, Sijben E, Bucur IG, Claassen T. Causal shapley values: exploiting causal knowledge to explain individual predictions of complex models.Adv Neural Inf Process Syst.2020;33:4778-89.Google Scholar Impellizzeri FM, Tenan MS, Kempton T, Novak A, Coutts AJ.急性:慢性工作量比:概念问题和基本误区。Int J Sport Phys Perform.2020;15(6):907-13.Article Google Scholar Harrell FE.回归建模策略:线性模型、逻辑回归和生存分析的应用》:Van den Goorbergh R, van Smeden M, Timmerman D, Van Calster B. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression.J Am Med Inform Assoc. 2022;29(9):1525-34.Article PubMed PubMed Central Google Scholar Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al.Epidemiol.2010;21(1):128.Article Google Scholar Dinga R, Penninx BW, Veltman DJ, Schmaal L, Marquand AF.超越准确性:评估机器学习模型的措施、陷阱和指南。BioRxiv. 2019;8:743138.Google Scholar Collins GS, Dhiman P, Navarro CL, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, van Smeden M. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence.BMJ Open.2021;11(7):e048008.Article PubMed PubMed Central Google Scholar Van Calster B, McLernon DJ, Van Smeden M, Wynants L, Steyerberg EW.校准:预测分析的致命弱点。BMC Med.2019;17(1):1-7.Google Scholar Collins GS, Altman DG.QRISK心血管风险预测的独立外部验证和评估:一项前瞻性开放队列研究。BMJ.2009;7(339):b2584.Article Google Scholar Collins GS, de Groot JA, Dutton S, Omar O, Shanyinde M, Tajar A, et al. External validation of multivariable prediction models: a systematic review of methodological conduct and reporting.BMC Med Res Methedol.2014;14(1):1-11.Google Scholar Bullock GS, Ward P, Impellizzeri FM, Kluzek S, Hughes T, Dhiman P, et al. The trade secret taboo: open science methods are required to improve prediction models in sports medicine and performance.Sport Med.2023;53:1-9.Article Google Scholar Download referencesNone.No funding was obtained for this letter.Authors and AffiliationsDepartment of Orthopaedic Surgery and Rehabilitation,Wake Forest University School of Medicine,Winston-Salem,NC,USAGarrett S.
Comment on: Machine Learning for Understanding and Predicting Injuries in Football
Dear Editor,
We recently read the article titled “Machine Learning for Understanding and Predicting Injuries in Football” in Sports Medicine – Open [1]. Given that injury prediction is an emerging topic within sport, the increasing interest and excitement towards complex machine learning algorithms within this space is a cause for concern when fundamental principles of prediction model development are not followed. As such, we feel the need to intervene and highlight methodological and conceptual inaccuracies.
The models presented in this paper were deemed by the authors to be “quite sound” [1]. However, this is not true, as recently highlighted in the systematic review in Sports Medicine [2]. All of these models were included in this systematic review, and after evaluation with the established Prediction Model Risk of Bias Assessment Tool (PROBAST) [3], were rated as high or unclear risk of bias [2].
The authors detail that, “the use of machine learning has great potential to unearth new insights into the workload and injury relationship.”[1] Prediction models may use both causal and non-causal predictors to estimate the risk of a future outcome [4, 5]. Consequently, it is inappropriate to use the included predictors to infer causal relationships between individual predictors and the outcome [6, 7]. Further, the authors state that Shapley values, local interpretable model-agnostic explanations, and partial dependency plots can be used to assist in interpreting cause-effect relationships with machine learning models [8]. These tools assess for associations between predictors and outcomes and, regardless of how these methods are labelled, the popular adage “correlation is not causation” still holds [8]. Importantly, these methods remain explorative, provide post hoc explanations (rationalization), and require confirmatory studies.
Such inaccurate and incorrect interpretations of clinical prediction models are of particular concern. This is because they can lead practitioners to attempt to change injury risk by intervening or manipulating predictor variables under the false assumption of a causal relationship; while these strategies are likely ineffective, they also have potentially harmful consequences for the athlete [4, 5, 9].
While the authors promote balancing dataset outcomes through over and under-sampling [1], this is highly discouraged as ‘balancing’ datasets alters the outcome prevalence, biasing towards overestimating risk.[10, 11] Balancing data without appropriate recalibration can inappropriately impact risk prediction and ultimately decision-making [11]. The authors also encourage creating classification models. Classification models are not recommended as this supersedes clinical and performance decision-making from the model users [11]. Classification models do not allow situational context and assume all situations and individuals have the same risk threshold. Prediction models should be developed and reported as a probability or at least risk score, to allow user interpretation and decisions [11].
The authors report that area under the curve (a form of discrimination), accuracy, sensitivity, and specificity should be used to evaluate machine learning models. These metrics are only a few of the recommended performance values that should be transparently reported [12] and some (such as accuracy) have well-known problems (also in machine learning) [13, 14]. The authors do not mention calibration as a model performance metric, which is the agreement between predicted and actual probabilities [15]. Calibration is always recommended and allows users to evaluate model performance across a range of risks [4, 5, 12]. Calibration divergence at specific risk ranges can alter clinical and performance decisions, even in models with high area under the curve performance [15].
The authors mention limitations of the included research, often only evaluating a single season of injuries for a given team, and how future research should seek to evaluate the models on subsequent seasons of data. Such a statement calls into question the notion of internal and external validation of these included papers. While the models in these papers have only been internally validated, their results provide a high level of optimism but no real understanding of how they may function under different circumstances or timepoints [16, 17]. Without a process of external validation for such models it is impossible to know whether or not they will be useful to practitioners with different clubs, observing different training constraints, under different coaches, and with different athletes who may have different base levels of risk [6, 18].
The purpose of this letter is to provide clarity about an important and burgeoning sports medicine and performance field. Other points that are not examined due to word count include sample size calculations, handling of missing data, data leakage, internal validation over train test splits, and clinical utility through clinical decision analysis and impact studies. Using the highest quality methods with transparency is imperative to protect and improve athlete health and performance, and to reduce the proliferation of conceptual and methodological inaccuracies about prediction model development and interpretation.
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
Majumdar A, Bakirov R, Hodges D, Scott S, Rees T. Machine learning for understanding and predicting injuries in football. Sports Med Open. 2022;8:73.
Article PubMed PubMed Central Google Scholar
Bullock GS, Mylott J, Hughes T, Nicholson KF, Riley RD, Collins GS. Just how confident can we be in predicting sports injuries? A systematic review of the methodological conduct and performance of existing musculoskeletal injury prediction models in sport. Sport Med. 2022;52(10):2469–82.
Article Google Scholar
Wolff RF, Moons KG, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Int Med. 2019;170(1):51–8.
Article PubMed Google Scholar
Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley R, Collins G. Methods matter: clinical prediction models will benefit sports medicine practice, but only if they are properly developed and validated. British J Sports Med. 2021;55(23):1319–21.
Article Google Scholar
Bullock GS, Hughes T, Sergeant JC, Callaghan MJ, Riley RD, Collins GS. Clinical prediction models in sports medicine: a guide for clinicians and researchers. J Orthop Sport Phys Ther. 2021;51(10):517–25.
Article Google Scholar
Bullock GS, Hughes T, Arundale AH, Ward P, Collins GS, Kluzek S. Black box prediction methods in sports medicine deserve a red card for reckless practice: a change of tactics is needed to advance athlete care. Sports Med. 2022;52(8):1729–35.
Heskes T, Sijben E, Bucur IG, Claassen T. Causal shapley values: exploiting causal knowledge to explain individual predictions of complex models. Adv Neural Inf Process Syst. 2020;33:4778–89.
Google Scholar
Impellizzeri FM, Tenan MS, Kempton T, Novak A, Coutts AJ. Acute: chronic workload ratio: conceptual issues and fundamental pitfalls. Int J Sport Phys Perform. 2020;15(6):907–13.
Article Google Scholar
Harrell FE. Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis: Springer; 2001.
van den Goorbergh R, van Smeden M, Timmerman D, Van Calster B. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression. J Am Med Inform Assoc. 2022;29(9):1525–34.
Article PubMed PubMed Central Google Scholar
Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for some traditional and novel measures. Epidemiol. 2010;21(1):128.
Article Google Scholar
Dinga R, Penninx BW, Veltman DJ, Schmaal L, Marquand AF. Beyond accuracy: measures for assessing machine learning models, pitfalls and guidelines. BioRxiv. 2019;8:743138.
Google Scholar
Collins GS, Dhiman P, Navarro CL, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, van Smeden M. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11(7):e048008.
Article PubMed PubMed Central Google Scholar
Van Calster B, McLernon DJ, Van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):1–7.
Google Scholar
Collins GS, Altman DG. An independent external validation and evaluation of QRISK cardiovascular risk prediction: a prospective open cohort study. BMJ. 2009;7(339):b2584.
Article Google Scholar
Collins GS, de Groot JA, Dutton S, Omar O, Shanyinde M, Tajar A, et al. External validation of multivariable prediction models: a systematic review of methodological conduct and reporting. BMC Med Res Methedol. 2014;14(1):1–11.
Google Scholar
Bullock GS, Ward P, Impellizzeri FM, Kluzek S, Hughes T, Dhiman P, et al. The trade secret taboo: open science methods are required to improve prediction models in sports medicine and performance. Sport Med. 2023;53:1–9.
Article Google Scholar
Download references
None.
No funding was obtained for this letter.
Authors and Affiliations
Department of Orthopaedic Surgery and Rehabilitation, Wake Forest University School of Medicine, Winston‑Salem, NC, USA
Garrett S. Bullock
Department of Biostatistics and Data Science, Wake Forest School of Medicine, Winston-Salem, NC, USA
Garrett S. Bullock
Centre for Sport, Exercise and Osteoarthritis Research Versus Arthritis, University of Oxford, Oxford, UK
Garrett S. Bullock
Seattle Seahawks, Seattle, WA, USA
Patrick Ward
Centre for Statistics in Medicine, University of Oxford, Oxford, UK
Gary S. Collins
Department of Health Professions Institute of Sport, Manchester Metropolitan University, Manchester, UK
Tom Hughes
Institute of Sport, Manchester Metropolitan University, Manchester, UK
Tom Hughes
School of Sport, Exercise, and Rehabilitation, University of Technology Sydney, Sydney, Australia
Franco Impellizzeri
Authors
Garrett S. BullockView author publications
You can also search for this author in PubMedGoogle Scholar
Patrick WardView author publications
You can also search for this author in PubMedGoogle Scholar
Gary S. CollinsView author publications
You can also search for this author in PubMedGoogle Scholar
Tom HughesView author publications
You can also search for this author in PubMedGoogle Scholar
Franco ImpellizzeriView author publications
You can also search for this author in PubMedGoogle Scholar
Contributions
GB, PW, and FM conceived the manuscript idea. GB, PW, and FM wrote the first draft of the manuscript. GB, PW, GC, TH, and FM critically revised the manuscript. GB, PW, GC, TH, and FM approved the final version of the manuscript.
Corresponding author
Correspondence to Garrett S. Bullock.
Ethics Approval and Consent to Participate
No ethics or consent approval were necessary for this letter.
Consent for Publication
Not applicable.
Competing interests
The authors declare that they have no competing interests with the content of this letter.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
Cite this article
Bullock, G.S., Ward, P., Collins, G.S. et al. Comment on: Machine Learning for Understanding and Predicting Injuries in Football. Sports Med - Open10, 84 (2024). https://doi.org/10.1186/s40798-024-00745-1
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s40798-024-00745-1
Share this article
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative