Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Longitudinal workload monitoring to keep athletes healthy and performing : conceptual, methodological,… Windt, Johann Dirk 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2019_february_windt_johann.pdf [ 2.75MB ]
JSON: 24-1.0375765.json
JSON-LD: 24-1.0375765-ld.json
RDF/XML (Pretty): 24-1.0375765-rdf.xml
RDF/JSON: 24-1.0375765-rdf.json
Turtle: 24-1.0375765-turtle.txt
N-Triples: 24-1.0375765-rdf-ntriples.txt
Original Record: 24-1.0375765-source.json
Full Text

Full Text

LONGITUDINAL WORKLOAD MONITORING TO KEEP ATHLETES HEALTHY AND PERFORMING: CONCEPTUAL, METHODOLOGICAL, AND APPLIED CONSIDERATIONS IN SPORTS INJURY AETIOLOGY    by Johann Dirk Windt A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF Doctor of Philosophy in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Experimental Medicine)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)      December 2018 © Johann Dirk Windt, 2018 ii  The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the thesis entitled: _____________________________________________________________________________   submitted by   Johann Windt    in partial fulfilment of the requirements for  the degree of Doctor of Philosophy           in Experimental Medicine            Examining Committee: Karim Miran-Khan, Department of Family Practice / School of Kinesiology    Co-supervisor  Tim Gabbett, Gabbett Performance Solutions        Co-supervisor  Ben C. Sporer, Department of Family Practice        Supervisory Committee Member  Bruno D. Zumbo, Faculty of Education, Measurement, Evaluation, and Research Methodology  Supervisory Committee Member  Mark Beauchamp, Faculty of Education, School of Kinesiology      University Examiner   Jordan Guenette, Department of Physical Therapy         University Examiner iii  Abstract Sports injury aetiology is a process in which internal and external risk factors contribute to an inciting event. The last decade has seen a rapid growth in research that identified how training and competition workloads relate to sports injury risk. My literature review highlighted that existing aetiology models do not describe how workloads contribute to injury. Furthermore, injury risk fluctuates on a time-scale in parallel to these workloads (e.g. daily), which creates several methodological and statistical challenges that have largely been ignored. If researchers are to understand how athletes’ workloads relate to injury risk, their conceptual aetiological models must be updated to incorporate training and competition workloads, and they should use appropriate statistical analyses.   After my literature review, I divide this dissertation into three parts. In Part 1 I discuss how workloads relate to injury. I present a novel workload—injury aetiology model, which expands on previous aetiological frameworks and details 3 ways that workloads contribute to injury: 1) exposing athletes to external risk factors and potential inciting events, 2) reducing injury risk through beneficial physiological changes, and 3) increasing injury risk through transient negative changes in athletes’ internal risk profiles. I then present mediation and moderation as potential causal approaches to understand how athlete risk factors interact with workload changes to alter injury risk.   In Part 2 I tackle the methodological challenges of analysing workload—injury data. I reviewed prospective cohort studies that reported intensive longitudinal data to analyse workload—injury data in team sports. I identified that few studies utilised statistical approaches that align with iv  theoretical aetiology models or addressed the methodological challenges associated with longitudinal data. My analysis leads me to recommend mixed modeling as one advance, and I exemplify how it can be used by studying how player unavailability affects player outputs.    In Part 3 I integrate the conceptual and methodological considerations into two applied settings. First, I describe how a methodological/mathematical concern (mathematical coupling) may influence applied practice (multifaceted player load management) and research (explicitly reporting calculations). Finally, I use mixed modeling to examine pre-season workload and in-season injury risk, controlling for athletes’ weekly workloads.  v  Lay Summary Participating in sport provides many health benefits. However, every minute a person participates in sport, they are exposed to potential injury risk.   The last two decades of sports injury research identified multiple ‘risk factors’ that contribute to sports injuries. These include athlete characteristics (e.g. strength, age, flexibility) and environmental characteristics (e.g. aggressive opponents, playing surface) that may protect or predispose individuals to injury. Until recently, how sport participation itself inherently changes an individuals’ injury risk in the next game or practice session has been largely absent.   In this dissertation I provide conceptual and methodological frameworks for how sport participation relates to an individual’s injury risk. Ultimately, understanding how sport participation contributes to injury and analysing these data more appropriately may inform future injury prevention efforts to keep more athletes healthy and performing.    vi  Preface I, Johann Dirk Windt, composed this dissertation in its entirety, with guidance and input from my supervisory team – Drs. Karim Khan, Tim Gabbett, Bruno Zumbo, and Ben Sporer. All major chapters within the dissertation body have been published in peer-reviewed journals, except chapter 5, which has been accepted and is in Press at BMJ Open. The details of each publication are included below.   A version of chapter 3 has been published in the British Journal of Sports Medicine. I conceived the idea for the manuscript and wrote the first draft, while Professor Gabbett contributed to the critical revision of the manuscript.  Windt J, Gabbett TJ. “How Do Training and Competition Workloads Relate to Injury? The Workload—Injury Aetiology Model.” British Journal of Sports Medicine, July 14, 2016, bjsports-2016-096040.  A version of chapter 4 has been published in the British Journal of Sports Medicine. I conceived the idea for this manuscript and wrote the first draft. All the other authors contributed to critical revision of the manuscript.  Windt J, Zumbo BD, Sporer BC, MacDonald K, Gabbett TJ. “Why Do Workload Spikes Cause Injuries, and Which Athletes Are at Higher Risk? Mediators and Moderators in Workload–Injury Investigations.” Br J Sports Med, March 4, 2017, bjsports-2016-097255.     vii  A version of chapter 5 is in Press at BMJ Open (Accepted September 4th, 2018). I conceived the idea for the paper through discussing longitudinal data analysis with Prof. Zumbo. I designed the search strategy with Dr. Clare Ardern and ran the search. Dr. Tim Gabbett and I reviewed all the identified systematic reviews and consensus statements to identify relevant papers. Dr. Clare Ardern and I identified original prospective cohort studies in the included systematic reviews/consensus statements that met our inclusion criteria. We also coded the articles according to the 50+ methodological criteria we established. I wrote the first draft of the manuscript. All authors were responsible for critical revision of the original manuscript and its revised version.  Windt J, Ardern CL, Gabbett TJ, Khan KM, Cook CE, Sporer BC, Zumbo BD. “Getting the Most out of Intensive Longitudinal Data: A Methodological Review of Workload—Injury Studies.” BMJ Open. In Press (Accepted September 4th, 2018).  A version of chapter 6 has been published in Journal of Science and Medicine in Sport. I was primarily responsible for the study design (in collaboration with Drs. Ekstrand and McCall), data analysis (in collaboration with Dr. Zumbo) and writing the first draft of the manuscript. All authors contributed to critical revision of the manuscript.  Windt, Johann, Jan Ekstrand, Karim M. Khan, Alan McCall, and Bruno D. Zumbo. “Does Player Unavailability Affect Football Teams’ Match Physical Outputs? A Two-Season Study of the UEFA Champions League.” Journal of Science and Medicine in Sport, August 24, 2017.    viii  A version of chapter 7 has been published in the British Journal of Sports Medicine. I simulated the data for this project and wrote the initial manuscript. Dr. Gabbett provided valuable input at the initial conception of the idea and throughout the manuscript revision process.  Windt, J, Gabbett TJ. “Is It All for Naught? What Does Mathematical Coupling Mean for Acute:Chronic Workload Ratios?” British Journal of Sports Medicine, May 28, 2018, bjsports-2017-098925.  A version of chapter 8 has been published in the British Journal of Sports Medicine. Data collection for this chapter was completed by Daniel Ferris and Dr. Tim Gabbett with the Gold Coast Titans. The chapter started with a discussion between Prof. Tim Gabbett and me where we wanted to know whether players who completed more of the pre-season were better able to handle in-season demands. I performed the analysis and wrote the first draft of this manuscript, with input from all authors to refine the paper.  Windt J, Gabbett TJ, Ferris D, and Khan KM. “Training Load--Injury Paradox: Is Greater Pre-season Participation Associated with Lower in-Season Injury Risk in Elite Rugby League Players?” British Journal of Sports Medicine, April 13, 2016, bjsports-2016-095973.   *Note my co-supervisor, Professor Karim Khan, is Editor in Chief of the British Journal of Sports Medicine. BJSM is the #1 ranked journal in the Thomson-Reuter journal category of Sports Science/Sports Medicine (of 81 journals, impact factor 7.867). He was not involved in handling of any of my papers at the British Journal of Sports Medicine.     ix  Statement of Research Ethics Approval: No ethics approval was sought or needed for Chapters 3 4, 5, or 7. Chapter 6 was approved by the UEFA research ethics board. Chapter 8 was approved by the Research Ethics Board of Australian Catholic University.   Use of Published Material and Active Voice The UBC Faculty of Graduate and Postdoctoral Studies does not specify whether doctoral dissertations are to be written in the first or third person, in the active or passive voice, and examples of both can be seen on cIRcle – UBC’s online dissertation repository. Similarly, the Experimental Medicine Program does not specify a preferred style. As the leading medical journals in the field (e.g. JAMA, Lancet, The BMJ) recommend first person, active voice, my committee and I agreed that I would follow these journals’ author instructions for this dissertation.   All previously published material in this dissertation is reproduced with permission from the relevant journals (British Journal of Sports Medicine for Chapters 3, 4, 7 & 8, BMJ Open for Chapter 5, Journal of Science and Medicine in Sport for Chapter 6). Therefore, these papers (making up Chapters 3-8 of this dissertation) include plural language for the author team – ‘we’ instead of ‘I’. These pronouns were not changed for this dissertation. As a result, the reader will find what will, on the surface, appear to be ‘inconsistency’ in the voice between Chapters 1-2 and 9 (written specifically for the thesis), and Chapters 3-8 (published/accepted manuscripts).   x  Table of Contents Abstract ......................................................................................................................................... iii Lay Summary .................................................................................................................................v Preface ........................................................................................................................................... vi Table of Contents ...........................................................................................................................x List of Tables ............................................................................................................................ xviii List of Figures ............................................................................................................................. xix List of Abbreviations ................................................................................................................. xxi Acknowledgements .................................................................................................................. xxiii Dedication ...................................................................................................................................xxv Chapter 1: Introduction ................................................................................................................1 1.1 Sports injury research foundations.................................................................................. 1 1.1.1 What is an ‘sports injury’? ........................................................................................ 2 Defining ‘cases’ in sports injury research ......................................................... 2 Injury severity .................................................................................................... 4 Injury classification ........................................................................................... 4 Injury diagnosis ............................................................................................. 4 Injury mechanism .......................................................................................... 4 Recurrent or new injury................................................................................. 5 1.1.2 The burden of sports injuries .................................................................................... 5 1.1.3 Foundations of sports injury prevention ................................................................... 7 Injury surveillance – understanding the extent of the problem ......................... 8 xi Injury aetiology – understanding why sports injuries occur .............................. 9 1.2 Foundations of workload monitoring............................................................................ 10 1.2.1 Physiological basis for athlete monitoring .............................................................. 11 1.2.2 Defining and quantifying workloads ...................................................................... 13 External Load .................................................................................................. 14 Internal Load .................................................................................................... 15 1.2.3 Purposes of athlete monitoring ............................................................................... 15 Planning, periodising and implementing training............................................ 15 Predicting performance .................................................................................... 16 Monitoring adaptation to training loads .......................................................... 17 1.2.4 Sport workloads and injury risk .............................................................................. 18 1.3 Measuring and analysing athlete workload – a key for research and applied sports science  ...................................................................................................................................... 19 1.3.1 Traditional measurement property views ................................................................ 19 1.3.2 Unified validity theory ............................................................................................ 20 Validity is about inferences, not about measures ............................................ 21 There are multiple sources of validity evidence .............................................. 21 Validity is an integrative evaluative judgement .............................................. 21 How unified validity theory may contribute to workload—injury research ... 22 1.4 Methodological and statistical limitations in athlete monitoring and injury prevention research ..................................................................................................................................... 23 1.5 Integrated athlete monitoring and injury research and practice .................................... 24 Chapter 2: Aims of the Dissertation ...........................................................................................26 xii  Chapter 3: How do training and competition workloads relate to injury? The workload—injury aetiology model .................................................................................................................28 3.1 Preamble ....................................................................................................................... 28 3.2 Introduction ................................................................................................................... 28 3.3 Why do athletes get hurt? Injury etiology models ........................................................ 29 3.4 Workloads and injuries – what do we know? ............................................................... 36 3.4.1 Defining and quantifying workloads ...................................................................... 36 3.5 Monitoring workloads in relation to performance, periodisation, and overtraining ..... 38 3.5.1 Total workloads and injuries ................................................................................... 41 3.5.2 Acute:Chronic workload ratio and injuries ............................................................. 42 3.5.3 Workload – injury paradox ..................................................................................... 44 3.6 An updated injury etiology model – incorporating the effects of workloads ............... 47 3.6.1 The triple role of workloads – ‘exposure’, ‘fitness’, and ‘fatigue’ ......................... 50 3.6.2 How do workloads influence injury risk? Implications of the updated model ....... 50 3.6.3 Model limitations .................................................................................................... 52 3.7 Where to from here – research implications ................................................................. 54 3.8 Summary ....................................................................................................................... 55 Chapter 4: Why do workload spikes cause injuries, and which athletes are at higher risk? Mediators and moderators in workload–injury investigations ...............................................56 4.1 Preamble ....................................................................................................................... 56 4.2 Introduction ................................................................................................................... 56 4.3 Domino or dimmer? Differentiating ‘mediators’ and ‘moderators’ ............................. 57 4.4 Not so simple, not so fast .............................................................................................. 59 xiii  4.5 What rewards await at the finish line after jumping the hurdles?................................. 60 Chapter 5: Getting the most out of intensive longitudinal data: A methodological review of workload—injury studies (Paper 3) ...........................................................................................61 5.1 Preamble ....................................................................................................................... 61 5.2 Introduction ................................................................................................................... 62 5.3 Methods......................................................................................................................... 64 5.3.1 Article selection ...................................................................................................... 64 5.3.2 Patient and public involvement ............................................................................... 66 5.3.3 Article coding and description ................................................................................ 66 5.3.4 Assessing how statistical models aligned with Collins’ threefold framework ....... 66 Collins’ component 1 – The theoretical models that underpin athletic workloads and injury risk (in brief) .................................................................................. 67 Collins’ component 2 – Temporal design / data collection ............................. 69 Collins’ component 3 – Statistical model ........................................................ 69 5.3.5 Evaluating statistical approaches ............................................................................ 70 5.3.6 Data synthesis approach .......................................................................................... 71 5.4 Results ........................................................................................................................... 71 5.4.1 Data collection ........................................................................................................ 75 Injury definitions ............................................................................................. 75 Subsequent or recurrent injuries ...................................................................... 76 Workload definitions ....................................................................................... 76 Measurement frequency .................................................................................. 77 Handling missing data ..................................................................................... 77 xiv  5.4.2 Statistical analysis and reporting in included articles ............................................. 77 Data binning/aggregation ................................................................................ 77 Analysis methods ............................................................................................. 77 Typical uses of statistical tools ........................................................................ 79 Justifications for statistical approaches ........................................................... 80 Addressing analysis assumptions and model fit .............................................. 80 5.4.3 Alignment of authors’ statistical models with theoretical model and temporal design challenges .................................................................................................................. 81 5.5 Discussion ..................................................................................................................... 82 5.5.1 Consideration #1 – Theoretical theme – multifactorial aetiology .......................... 83 5.5.2 Consideration #2 – Theoretical theme - between and within-athlete differences ... 86 5.5.3 Consideration #3 – Theoretical theme - injury risk as a complex dynamic system 87 5.5.4 Consideration #4 – ILD challenge – including time-varying and time-invariant variables ................................................................................................................................ 88 5.5.5 Consideration #5 – ILD challenge – handling missing and unbalanced data ......... 89 5.5.6 Challenge #6 – ILD challenge – dependencies created by repeated measures ....... 90 5.5.7 Challenge #7 – ILD challenge – incorporating time into the analysis/temporality 92 5.5.8 Methodological, statistical and reporting considerations........................................ 93 Data aggregation .............................................................................................. 93 Checking model assumptions and fit ............................................................... 94 5.5.9 Researcher ‘trade-offs’, consequences of misalignment ........................................ 94 5.5.10 Review limitations .................................................................................................. 95 5.5.11 Longitudinal improvements in ILD analysis .......................................................... 96 xv  5.5.12 Future directions and recommendations for ILD analysis ...................................... 98 5.6 Conclusion .................................................................................................................. 101 Chapter 6: Does player unavailability affect football teams’ match physical outputs? A two-season study of the UEFA champions league (Paper 4) ..................................................103 6.1 Preamble ..................................................................................................................... 103 6.2 Introduction ................................................................................................................. 104 6.3 Methods....................................................................................................................... 105 6.4 Results ......................................................................................................................... 108 6.5 Discussion ................................................................................................................... 114 6.6 Summary and conclusions .......................................................................................... 116 Chapter 7: Is it all for naught? What does mathematical coupling mean for acute:chronic workload ratios?  (Paper 4) .......................................................................................................118 7.1 Preamble ..................................................................................................................... 118 7.2 Introduction ................................................................................................................. 119 7.3 Comments on mathematical coupling and ACWRs: .................................................. 120 7.4 What now? A call for critical, thoughtful monitoring ................................................ 125 Chapter 8: A training load--injury paradox: Is greater pre-season participation associated with lower in-season injury risk in elite rugby league players? (Paper 6) ...........................127 8.1 Preamble ..................................................................................................................... 127 8.2 Introduction ................................................................................................................. 128 8.3 Methods....................................................................................................................... 129 8.3.1 Study design .......................................................................................................... 129 8.3.2 Measures ............................................................................................................... 129 xvi Injury status ................................................................................................... 130 Pre-season attendance: ................................................................................... 130 Quantifying in-season training loads ............................................................. 130 8.3.3 Data collection and analysis procedures ............................................................... 131 8.3.4 Statistical analysis ................................................................................................. 131 8.4 Results ......................................................................................................................... 132 8.4.1 Injuries .................................................................................................................. 132 8.4.2 Pre-season participation ........................................................................................ 133 8.4.3 Pre-season loads and injury risk ........................................................................... 133 8.4.4 Training loads and injury risk ............................................................................... 135 8.4.5 Full injury prediction models ................................................................................ 139 8.5 Discussion ................................................................................................................... 142 8.5.1 Training loads and injury likelihood ..................................................................... 142 8.5.2 Pre-season participation and in-season injury – providing protection or revealing underlying differences [i.e. identifying the robust players]? .............................................. 143 8.5.3 Preparation through pre-season participation – practical implications ................. 144 8.5.4 Potential limitations .............................................................................................. 144 8.5.5 Summary and conclusions .................................................................................... 145 Chapter 9: Conclusion ...............................................................................................................146 9.1 Integrating my findings into the field, and contributions of my dissertation.............. 146 9.1.1 Conceptual contribution ........................................................................................ 146 9.1.2 Methodological contributions ............................................................................... 149 9.1.3 Applied contributions............................................................................................ 151 xvii  9.2 Dissertation strengths .................................................................................................. 153 9.3 Dissertation limitations - what I would do differently ................................................ 153 9.4 Future research directions ........................................................................................... 154 Bibliography ...............................................................................................................................158 Appendices ..................................................................................................................................183 Appendix A - Additional tables and figures from Chapter 5 .................................................. 183 Appendix B - Additional tables and figures from Chapter 6 .................................................. 190  xviii  List of Tables Table 3.1: Measures of external and internal load ........................................................................ 38 Table 3.2: Relationships between injury risk and various acute:chronic workloads combined with high and low chronic workloads ................................................................................................... 46 Table 5.1: Summary of included workload—injury investigations, sorted by sport then publication year ............................................................................................................................. 72 Table 5.2 Broad injury definitions used in workload—injury investigations ............................... 75 Table 5.3: Independent variables used in workload—injury investigations ................................. 76 Table 5.4: The number of studies using various statistical analysis techniques ........................... 78 Table 5.5: Evaluation of the degree to which authors’ use of statistical tools addressed theoretical and temporal design challenges .................................................................................................... 85 Table 6.1: Average and variability of teams’ physical outputs in the 2014-2015, 2015-2016 UCL seasons ........................................................................................................................................ 110 Table 7.1: Coupled and uncoupled (rolling averages over 4 weeks) ACWR calculations ......... 122 Table 8.1. Descriptive statistics for players’ average workload and injuries over the duration of the study. ..................................................................................................................................... 135 Table 8.2 Association of training load variables with injury likelihood in the current and subsequent week. ........................................................................................................................ 138 Table 8.3. Effect of pre-season participation on injury likelihood in current and subsequent week, controlling for training load variables. ........................................................................................ 139    xix  List of Figures Figure 1.1: Proposed model of the relationship between external and internal training loads. .... 14 Figure 3.1: Meeuwisse’s first multifactorial injury etiology model proposed in 1994. ............... 31 Figure 3.2 Bahr & Krosshaug’s 2005 comprehensive etiology model. ........................................ 33 Figure 3.3: Meeuwisse and colleagues’ 2007 dynamic, recursive sports injury etiology model. 35 Figure 3.4: Injury likelihood associated with various weekly changes in workloads. ................. 43 Figure 3.5: Finding the ‘sweet spot’ for acute:chronic workload ratios in return-to-play protocols........................................................................................................................................................ 44 Figure 3.6: The workload – injury etiology model. ...................................................................... 48 Figure 4.1: Examples of how mediating and moderating variables explain the association between a workload spike and subsequent injury. ........................................................................ 58 Figure 5.1: Complex systems model of athletic injury. ................................................................ 68 Figure 6.1: Teams’ average player unavailability for each season and their average: 1) Distance per game (m), 2) Number of sprints (efforts >20km/h) per game, and 3) Percent of total distance covered above 14km/h per game. ............................................................................................... 112 Figure 6.2: Estimated effects of 4 additional unavailable players on physical output measures 113 Figure  7.1: The distinction between traditional ‘coupled’ and ‘uncoupled’ estimates of the acute:chronic workload ratio....................................................................................................... 120 Figure 7.2: The association between coupled and uncoupled estimates of the acute:chronic workload ratio with increasing workload progression. ............................................................... 124 Figure 8.1: The association of pre-season participation with games missed due to injury. ....... 134 Figure 8.2: Weekly training loads and injury incidence ............................................................. 136 xx  Figure 8.3: Predicted injury probabilities based on high speed running percentage and pre-season participation. ............................................................................................................................... 141      xxi  List of Abbreviations  ACL: Anterior Cruciate Ligament ACWR: Acute:Chronic Workload Ratio AFL: Australian Rules Football League AIC: Aikeke Information Criteria AMPK: 5' Adenosine Monophosphate-Activated Protein Kinase ANOVA: Analysis of Variance BIC: Bayesian Information Criteria CI: Confidence Interval DAG: Directed Acyclic Graph DALDA: Daily Analysis of Life Demands UEFA: Union of European Football Associations  GEE: Generalised Estimating Equation GPS: Global Positioning Systems HLM: Hierarchical Linear Modeling HR: Heart Rate ILD: Intensive Longitudinal Data IOC: International Olympic Committee MAR: Missing at Random MBI: Magnitude Based Inferences MCAR: Missing Completely at Random mTOR: Mammalian Target of Rapamycin NFL: National Football League xxii  OR: Odds Ratio OSTRC: Oslo Sports Trauma Research Centre POMS: Profile of Mood States PROM: Patient-Reported Outcome Measure REST-Q: Recovery Stress Questionnaire RR: Relative Risk sRPE: session Rating of Perceived Exertion TRIMP: Training Impulse TRIPP: Translating Research into Injury Prevention Practice  TSB: Training Stress Balance UCL: UEFA Champions League UEFA: Union of European Football Association  xxiii  Acknowledgements Professor Karim Khan, words cannot convey how much of a privilege it has been to work with you these last 5+ years, from a Master’s degree through to the completion of this doctorate program. When I asked you in the winter of 2013 whether you would recommend I tackle an MD or PhD, you suggested that I start with an MSc with you. I would never have guessed how it would change my life. Like most graduate students, I have learned about research, writing (verbs trump nouns), and academic life while working with you. However, all these lessons seem so small when compared to the lessons of leadership, collaboration, management, culture, and teamwork. Countless people have deeply benefitted from your mentorship and desire to see others succeed, and I am thankful to be among them.   Professor Tim Gabbett, you have literally changed the vocabulary of an entire field of research and applied practice. The words ‘acute’ and ‘chronic’ will never be the same! More importantly, seeing you integrate your research into helping real athletes in the field is inspiring. Most of all, I will always remember that while “what we do is important, none of it matters”, when compared to the joys of family, friends, and faith.   Dr. Ben Sporer, I will never use a measure, integrate a new technology, or take on a project without being able to clearly answer the question “What are you trying to achieve?”. You have challenged my thinking, inspired my work, and continually drawn me back to the foundational physiological principles upon which we build our interventions to keep athletes healthy and performing their best.  xxiv  Professor Bruno Zumbo, I am so thankful that our paths crossed early on in my doctoral degree and that our meeting has spawned a fruitful and ongoing collaboration and friendship. You are a mentor to so many in academia and life. I am honoured to be one of those on the receiving end.   I have been generously supported by the Canadian Institutes of Health Research, receiving a Vanier Scholarship for my PhD work, which enabled me to focus on my research endeavours during this time.  To my loving parents, Adriaan and Henriette Windt, I owe you both an incredible debt of gratitude. You have lavished your love, time, encouragement, and financial support throughout each stage of my life and academic journey. I don’t know if I could ever quite convey how thankful I am for you both, but I hope and pray I will make you proud and demonstrate how much I love you.  Most of all, to my incredible wife, Sheralyn. You have loved me, supported me, encouraged me, and empowered me – more than I could ever have hoped for. Words cannot express how blessed I am to have you. I love you and thank you from the bottom of my heart. As this chapter closes and we embark on a new journey together as parents, I am excited to spend our evenings and weekends with our daughter at the park instead of doing grad school at a coffee shop.   xxv  Dedication Everly, my beloved daughter – I love you.    To my Lord, my wife, my family and friends. 1  Chapter 1:  Introduction While sport provides a means of physical activity and is thereby associated with a plethora of health benefits [1,2], participation comes with an inherent injury risk. While certain injuries occur more frequently in the presence of certain risk factors – e.g. higher anterior cruciate ligament (ACL) injury risk in female athletes [3] – all sports injuries occur during training or competition. Beyond simple ‘exposure’ to injury risk, the recursive nature of injury risk means that each training or competition event may alter individuals’ subsequent risk [4]. To understand the integral role of training and competition workloads in sports injury development, it is necessary to understand both 1) sports injury prevention and 2) workload monitoring. Once I have summarised these foundations, I will describe (i) how these two fields interact, (ii) the methodological and statistical challenges that have limited progress to date, and (iii) how both conceptual and methodological considerations may impact applied practice and research.   1.1 Sports injury research foundations Sports injury research begins with a fundamental question: what constitutes a sports injury? After defining ‘sports injury’, I will describe the burden that sports injuries pose at the public health, organisation, and individual athlete levels. I will then outline the principles of injury prevention through the lens of Professor Willem van Mechelen’s ‘sequence of prevention’ model [5].     2  1.1.1 What is a ‘sports injury’? Injuries occur when the transfer of physical forces exceed the body’s ability to handle them, and structural damage occurs [6]. While acknowledging that differentiating ‘disease’ and ‘injury’ presents several theoretical challenges [7], the above definition provides a basis to ground sport injury discussions. It also differentiates musculoskeletal sports injuries from other sports-related conditions, such as sports-related asthma or  cardiac events. To operationalise the definition, I will outline the key characteristics of sports injury research – the definition, severity, mechanism, and classification of injuries. Defining ‘cases’ in sports injury research  Injury definition is vital in the science of epidemiology [8]. Prior to 2005, few efforts were made to reach consensus on sports injury definitions [6]. A landmark sports injury definition was published in 2006 following the 1st World Congress on Sports Injury Prevention, where an expert panel defined a football injury as [9]:  “Any physical complaint sustained by a player that results from a football match or football training, irrespective of the need for medical attention or time loss from football activities. An injury that results in a player receiving medical attention is referred to as a ‘medical attention’ injury, and an injury that results in a player being unable to take part in future football training or match play as a ‘time-loss’ injury.” [9]  Other sports (e.g. cricket, rugby league, rugby union) have since adopted similar consensus definitions [10–15]. Within these frameworks, injuries may be divided into 3 separate categories.   3   1) Any physical complaint - sustained by a player that results from a football match or football training, irrespective of the need for medical attention.  2) Medical attention injury – any injury that requires medical attention 3) Time loss injury – any injury that causes a player to lose training or match time.   Researchers citing the consensus document use the ‘time loss’ injury definition almost exclusively [16,17]. They argue that time loss, and even match-time loss definitions [18], are the most appropriate sport injury definitions because they 1) impact team performance and are therefore most relevant to coaching staff, and 2) can be collected more readily (which makes it feasible when resources are limited).  However, time loss is the least sensitive measure of sports injury and it captures fewer injuries than the other two measures of injury [16]. Norway’s Professor Roald Bahr raised the argument that time loss measures underestimate the true burden of injuries in sports like volleyball, swimming, and high jumping, where athletes report pain and decreased function characteristic of ‘overuse’ injuries but may not miss training or competition time. In those sports, time-loss definitions provide low injury incidence, but the prevalence of pain is high and this can compromise player performance [16]. To capture this type of injury, Professor Bahr and colleagues, including Dr Ben Clarsen, developed the Oslo Sports Trauma Research Centre (OSTRC) Overuse Injury Questionnaire  [19,20]. The key elements of this patient rated outcome measure (PROM) questionnaire included the weekly reporting of injuries, illnesses, or other health problems, and their effect on athletes’ training volume, performance, or symptoms [19,20].   4 Injury severity Major consensus statements define injury severity as “the number of days that have elapsed from the date of injury to the date of the player’s return to full participation in team training and availability for match selection” [9]. Grouping injury severity into categories like – slight (0 days) minimal (1 to 3 days), slight (4 to 7 days), moderate (8-28 days), severe (>28 days), and career ending is also common practice and described in the original football consensus statement [9]. Injury classification Injuries can be classified according to their anatomical diagnosis, mechanism (overuse, acute), and whether they are new or recurrent. Injury diagnosis Classifying injuries according to their diagnosis includes identifying the general body region injured, the structures involved, and the specific injury nature. Traditional medical classification methods such as the International Classification of Diseases (ICD) coding system [21,22] and others [23] have major limitations when classifying sports injuries [24,25], so several sport specific coding systems have been developed. The Sports Medicine Diagnostic Coding System (SMDCS) [26], and the Orchard Sports Injury Classification, now in its 10th version [27–29], are both widely used in sports injury epidemiology settings [30–33]. Injury mechanism Sports injuries may have an acute onset (i.e. usually to trauma) or overuse (i.e. gradual) onset. Acute injuries are those with a sudden, identifiable onset, often traumatic in nature [34]. Examples 5  of acute injuries include ACL ruptures, ankle ligament sprains, and concussions. Overuse injuries, as their name suggests, are generally described as accumulating from excessive strain over time. Their onset is progressive and the injury ‘event’ is difficult to identify [35]. Describing the inciting event provides greater understanding of the injury mechanism, such as whether it was contact or non-contact in nature [36]. Recurrent or new injury Recurrent injuries are defined in consensus statements as injuries of the same type and at the same site as an original/index injury, occurring after an athlete returns to full participation [9]. Following these consensus statements, recurrent injuries were divided into ‘exacerbations’ if athletes had not returned to full participation, and ‘re-injuries’ if they had returned to full participation [37]. Although it is beyond the scope of this dissertation, accurately classifying recurrent injuries has received much attention in the last decade because previous injury is the most commonly cited risk factor for future injury, and understanding re-injury or injury exacerbation risks wholly depends on accurate classification [38–44].  Epidemiology relies on accurate case definition. Similarly, these considerations (injury definition, severity, and classification) are critical in sports injury research.    1.1.2 The burden of sports injuries  Sports injuries pose a notable public health burden, especially in youth populations [45]. The total economic burden of injuries in Canada is estimated at over $26 billion [46], and sports injuries make up 35% of injuries [47]. Youth (aged 12-19) are the most likely Canadian age group to 6  experience an injury, and sport causes 2 out of 3 injuries within this demographic [47]. These trends resemble those documented by international surveys [48,49]. Instead of surveys, medical records or insurance billing data may also be used to understand the public health burden of sport injuries. However, although sports injuries are a major cause of emergency room visits, many do not result in hospitalisation, so population-level data using these data are likely underreported [50].   Narrowing the scope from the population level to a team level, sports injuries carry performance and financial ramifications. Team sport success is compromised by high injury rates [51–53]. In professional sport, the financial costs associated with time loss injuries are substantial. For example, a single hamstring strain injury in Australian Rules Football (AFL) players was costed at over AU$ 40,000 (Australian dollars) because of loss of the player’s time from his employment [54]. In 2012, the average financial cost per AFL team exceeded AU$ 240,000 for hamstring strain injuries alone. These financial costs are even greater in sports where salaries are greater– e.g. a month of time-loss for a European soccer player is estimated at €500,000 [55], and hamstring strains cost US$350,000 per National Football League (NFL) team [54].   Narrowing the scope one step further, injuries burden individual athletes through performance decrements. Systematic review evidence demonstrates that injuries impair the chances of successful performance in athletes [56]. Elite track and field athletes were 7 times more likely to reach their predetermined performance goals if they were able to complete greater than 80% of their planned training [57]. Each missed week of training reduced the odds of success by 26% [57].  Depending on the injury, players may be out of competitive participation for weeks or months, and potentially encounter physical, financial, and psychological challenges associated with their 7  injuries. Once they return to sport, they may remain at increased risk for subsequent injury because previous injury is considered a risk factor for future sports injuries [58]. Sporting injuries like anterior cruciate ligament can have a long-term health impact on players, even after they discontinue their sporting participation [59–61]. Finally, sport injuries have also been linked to increased opioid prescription and misuse after athletes’ careers end [62].   1.1.3 Foundations of sports injury prevention Professor van Mechelen’s 1992 landmark ‘sequence of sports injury prevention research’ model outlined a 4-step process: 1) injury surveillance to establish the extent of the injury problem, 2) aetiological research to identify relevant risk factors, 3) preventative strategies to correct modifiable risk factors, and 4) re-evaluating the injury problem to assess a preventative measure’s effectiveness [5]. Over a decade later, Prof. Caroline Finch identified that van Mechelen’s model was limited by its reliance on the scientific environment; it overlooked the need to implement preventative measures in the real world. She introduced the Translating Research into Injury Prevention Practice (TRIPP) Framework [63]. Keeping the first 3 steps from van Mechelen’s model, Prof. Finch changed the 4th stage to scientific evaluation of preventive measures in ‘ideal conditions’, introduced a 5th stage – describing intervention context to inform implementation strategies – and a 6th stage – evaluating effectiveness of preventive measures in the implementation context. My dissertation is primarily on sports injury aetiology, so I will outline the first two steps in van Mechelen’s model below before proceeding to understanding athlete workloads.  8 Injury surveillance – understanding the extent of the problem Sports injury surveillance is systematically tracking injury occurrences within a sporting population across a given time. Surveillance is not an end in itself, but it is a necessary step to establish the incidence of injury across sporting contexts, and at different times within a sporting environment [63].  While injury incidence (injury cases/1000 participation hours) is the most common outcome of injury surveillance studies, it is not the only one. For example, prevalence proportion (usually point prevalence) equals the number of injured athletes at a given time divided by the total number of athletes [64]. In sports where overuse/chronic injuries are common and athletes may play through pain, some have argued that prevalence measures may be more appropriate than incidence measures [16,65]. While injury incidence is usually expressed as an injury incidence rate (most common – e.g. injuries/1000 hours), incidence proportions are also possible – total # of injured athletes over a given time (e.g. season) divided by the total number of athletes [64]. Accurate surveillance should also include injury severity – the number of days lost due to an injury. Finally, injury burden may be calculated by multiplying injury incidence by injury severity, which may identify the injuries with the most substantial impact [66,67].  Injury surveillance also provides the evidence to evaluate injury risk across and within various sports settings [63]. Different physiological demands, technical demands, and playing characteristics create disparate injury risk across sports. These differences also extend to different competitive levels (i.e. recreational vs. professional), and sporting activity (i.e. training vs. competition).   9 Injury aetiology – understanding why sports injuries occur Aetiology is the study of why events occur. Building on research from the field of infectious diseases [68], Canada’s Professor Winne Meeuwisse first described sports injuries as multifactorial in 1994 [69]. His model described how internal (e.g. age, sex, flexibility) and external risk factors (e.g. playing surface, opponent behaviour, equipment) contribute to risk and may eventually coincide in an inciting event [69]. Meeuwisse subsequently extended this model and clarified that these risk factors may interact or confound the relationship between other risk factors and injury risk [70].  Two subsequent iterations of the multifactorial injury aetiology model expanded on methodological considerations [71] and defined the inciting event details [36]. In 2007, Meeuwisse and colleagues described the recursive nature of sports injury risk [4].   Accurate injury surveillance that tracks injury events and measures multiple risk factors can identify subgroups that are at higher or lower injury risk. Risk factors may be modifiable or non-modifiable. Using ACL injuries as an example – women’s football (soccer) players may be at a fourfold higher risk compared to their male counterparts [3]. Among female athletes, those with higher knee abduction moments and knee abduction angles during landing are at an increased risk of ACL tears [72]. Sex exemplifies an essentially non-modifiable risk factor while neuromuscular control during landing exemplifies a modifiable factor [73].   Training and competition workload –the intensity and duration of sporting activity – has been increasingly linked to injury risk [74–76]. While workload may be considered a risk factor for 10  sports injury, it has not been included in previous aetiological models, and whether it should be classified as an internal or external risk factor remains unknown.   1.2 Foundations of workload monitoring As sport has prehistoric origins, so too does training for sports and hence ‘workload’ – the volume and intensity of training and competition in sport. Workload measurement forms one substantial pillar of my PhD research. Research interest in workload has exploded in the last decade because of inventions and advances in technologies such as Global Positioning Systems (GPS), accelerometers, heart rate monitors, video tracking systems, and others, all of which allow athletes and coaches to quantify workload in much more detail than they could even 20 years ago. Training load monitoring to measure athletes’ training, recovery, performance, and stress has become common practice [77]. Statistical advances and the computational power to manage vast datasets have further created what we might consider a niche in the applied sports sciences.  The remainder of part 1.2 is my concise summary of the key literature in this field that has informed my dissertation.   Training load is a combination of training intensity, duration and frequency” (Smith, 2003, pg. 1111). Training load monitoring was rooted in observing athletes during workouts – particularly endurance athletes’ heart rates during steady state or interval training or performing index/standard workouts [79].    11  1.2.1 Physiological basis for athlete monitoring General adaptation syndrome (also widely know as  the “stress response”) was described in 1936 by Canadian scientist Hans Selye, who described three phases after a stressor was applied: 1) alarm, 2) resistance, and 3) exhaustion [80,81]. Although challenged in the literature [82,83], the exercise sciences rely heavily on Selye’s initial work in describing the physiological adaptations to training and subsequent ‘supercompensation’ [84,85]. A training ‘stress’ induces a period of alarm (acute fatigue). This is followed by a period of resistance wherein the body adapts to the stress so that it is better equipped to handle the stressor in the future (i.e. supercompensation). In the sport and exercise sciences, Selye’s work has been adopted an informed modern-day periodisation theory [84,86].  Periodisation theory dictates that subsequent stressors should be applied after enough time has elapsed for supercompensation. Waiting too long between stressors means the athlete loses adaptations and returns to a baseline level (detraining), while too little time between stressors may interrupt the adaptation process. If high volume and intensity training is continually repeated without adequate rest, maladaptation and overtraining may result [87–89], and this is akin to Selye’s exhaustion.   I would respectfully contend that, as written, general adaptation syndrome may oversimplify the training load—response process in two ways. First, adaptation occurs at the cellular/molecular level [90–92]. At the cellular level, exercise disrupts homeostasis and activates several signals, including mechanical stretch, calcium concentrations, hypoxia, and muscle energy status. Based on the strength and combination of these signals, signalling cascades lead to transcription and 12  translation of proteins, which will facilitate improved capacity to handle similar stressors in the future [92,93]. Two primary examples include 1) resistance training activating the mTOR pathway, which leads to muscle hypertrophy through satellite cell proliferation, and 2) endurance training activating the enzyme best know as AMPK (5' AMP-activated protein kinase) , which facilitates mitochondrial biogenesis and consequently improves cellular aerobic capacity [92]. Therefore, contextualising the ‘stress’ and ‘adaptation’ to the specific training goal extends a ‘general’ view of adaptation.   The second oversimplification from general adaptation syndrome is assuming a consistent, solely biologically mediated stress—response relationship. This assumption is challenged by the individual variability to standard training programs [94]. As the science into the elements of stress has advanced [95], Dr. John Kiely has challenged traditional periodisation theory [82,83]. Contrary to periodisation plans that assume a purely biologically mediated, consistent dose-response relationship, Kiely contends that a given workload will produce varying adaptations, depending on an athletes genotype, phenotype, stress history, prior training history, and current stress status (e.g. psycho-emotional and cognitive state) [83]. Acknowledging these oversimplifications, general adaptation syndrome underpins modern training principles. Therefore, coaches, athletes, and of course researchers, are motivated to quantify athletes’ training and the way they are (or are not) responding. In short, ‘workload’ is the main modifiable variable that affects performance, but the relationship is complex. Coaches seek to measure the various dimensions of workload (see below) accurately and titrate workload to try to optimise the athlete’s performance.    13  1.2.2 Defining and quantifying workloads   ‘Load’, according to a recent International Olympic Committee (IOC) consensus statement [65], is the “sport and non-sport burden (single or multiple physiological, psychological, or mechanical stressors) as a stimulus applied to a human biological system” (pg. 1031). In this dissertation, I will use this definition for ‘load’ and ‘workload’ interchangeably. Using this definition, workloads can also be differentiated into external and internal training loads. External workloads refer to the work performed by the athlete, while the internal workload describes athlete’s physiological response to a training bout [96]. A given external load will elicit different internal responses in each individual, based on external load characteristics (quality and quantity/organisation), as well as the athlete’s characteristics (genetics, fitness level, training background, etc.) (Figure 1.1) [97]. Careful consideration should be given to what internal or external measures of load are most appropriate, valid, and reliable given the sporting/athlete context, and the goals in collecting these data [77].   14   Figure 1.1: Proposed model of the relationship between external and internal training loads. With permission, from [97]. External Load  External load refers to what an athlete ‘does’, measured using objective criteria. The simplest examples include total distance run (e.g. meters) or weight lifted (e.g. kg) but today, external load can include power output (Watts), speed (e.g. m/s), acceleration/deceleration (m/s2), and neuromuscular function (e.g. sprint performance (s), countermovement jump test (rate of force development in Newtons/second, jump height in cm). The widespread adoption of wearable microsensors, which use global position system (GPS) and accelerometers to collect load data have become the most common means of collecting external load in many sporting contexts [98,99]. 15 Internal Load  Internal load refers to an athlete’s physiological or perceptual response to a given external load. An athlete group could be given a standard external load – e.g. a 2km run in 10 minutes, but their internal loads – e.g. heart rate, lactate accumulation, rating of perceived exertion (RPE) would all differ. There is some debate about whether psychological variables (e.g. Profile of Mood States), which capture an athlete’s current psychological state, should also be considered internal load measures [77,96]. I respectfully contend, with other applied workload—injury researchers, that psychological variables should be analysed separately within the ‘workload-monitoring cycle’, because they represent an athlete’s subjective wellness that culminates through training and non-training stressors [100].   1.2.3 Purposes of athlete monitoring Athlete monitoring can serve at least 4 purposes; 1) Planning, periodising, and implementing training plans, 2) predicting performance, 3) monitoring adaptation, and 4) mitigating injury risk. Planning, periodising and implementing training Periodised training programs attempt to apply training loads with specific proportions and magnitudes to maximise the benefits of adaptation [101]. These programs may err in two ways: 1) training loads that are too low fail to promote adaptation, or 2) training loads that are too high increase the potential for maladaptation, illness, and overtraining, as well as increased injury risk [101]. I appreciate that this ‘Goldilocks principle’ for training is easy to articulate and difficult to 16  execute – my thesis aims to contribute to execution, particularly through mitigating injury risk as one potentially negative outcome.  Monitoring training loads facilitates the implementation of periodised plans in at least three ways. First, coaches’ planned training loads and actual athlete training loads may not match, so measuring player workloads allows coaches to understand actual loads. [102]. Second, load monitoring can be used to track how athletes are performing and responding to training over time. By understanding external loads, internal loads, athlete wellness and readiness, practitioners can make on-the-fly modifications to account for individual differences in training responses [82,100]. Finally, elite athletes are differentiated by extremely small margins [103]. Tapering is commonly used to minimise fatigue and maximise performance at key events [104]. However, to perform an effective taper and achieve the performance benefits, loads need to be accurately and reliably tracked, so that they can be precisely altered leading into competition. There is room for more science in this field. Predicting performance In the mid-1970s, Professor Eric Banister and colleagues proposed a systems model for predicting performance following training, which built on the foundation of the positive and negative training effects [105]. Their model included a positive function of training effect (fitness), and a negative function for fatigue based on an athlete’s heart-rate based training impulse (TRIMP). An athlete’s modeled performance was the difference between fitness and fatigue. Subsequent performance prediction models [106,107] mathematically model training loads and expected responses.  While criticisms have been leveled against Banister’s model since it oversimplifies the training—17  response relationship [108], these models all analysed longitudinal load data to predict future performance and provided coaches/scientists a framework to understand athlete training responses [77]. Monitoring adaptation to training loads By monitoring training loads with health and performance measures, researchers can assess an athlete’s fatigue levels and his/her potential readiness to train. Improved competition performances or maximal exercise tests may indicate training adaptation, but also cause substantial fatigue on the athlete and are therefore used sparingly. Therefore, less fatiguing methods are regularly utilised. Individual deviation from ‘normal’ loads during similar load sessions, or discrepancy between internal and external load measures, could indicate a state of fatigue [96]. These training load measures can be used in conjunction with other markers, such as heart rate variability, hormone levels (testosterone, cortisol, catecholamines), mood state and recovery status (REST-Q, POMS), and others [109].  Submaximal ‘index’ tests have also been designed for similar purposes [110].    In addition to ‘objective’ measures of performance, athlete self-report (e.g. wellness surveys) can provide a simple measure of athlete response to training. Dr. Anna Saw and colleagues concluded in a 2016 systematic review that these were superior to the objective tests of athlete responses.  [111]. As there is no single ‘best’ marker of training response, current gold standard among professional teams is to collect several markers of training load and athlete responses.    18  1.2.4 Sport workloads and injury risk  In the last 15 years, a growing body of research has investigated the links between training workloads and sports injuries. Systematic reviews [74–76] and a 2016  International Olympic Committee (IOC) Consensus Statement [65] have found that rapid workload increases are associated with increased injury risk, while findings regarding high absolute workloads have differed – some showing them as a risk factor, while others identifying them as protective. Australia’s Dr. Tim Gabbett’s ‘training load injury prevention paradox’ paper [112] described how high chronic workloads may be protective against injury, especially when athletes get to those loads safely [113,114].   Surprisingly, training and match workloads had been overlooked in the aetiological models [4,36,69,71] that existed when I started this thesis. Workloads are unlikely to contribute to all injuries, since each injury stems from a unique collection of risk factors. For example, workload likely plays a lesser role in an ice hockey player’s concussion sustained from an illegal cross-check to the head than a soccer player’s hamstring strain while sprinting for the ball. Therefore, a number of workload—injury investigations focused on non-contact soft-tissue injuries as the primary outcome, since these are considered ‘preventable’ and may be more sensitive to load changes [115]. To summarise this section, I note that there has been no synthesis of traditional risk factors along with workload in previous aetiology models. (Please see Chapter 3)  19  1.3 Measuring and analysing athlete workload – a key for research and applied sports science To date this literature review has focused on workload as a concept.  In practice, what is the best measure of workload? Professional teams take various factors into account including cost, technological capabilities, cultural concerns (will athletes agree to wear heart rate monitors), and personnel. I contend that the measurement properties of any workload metric must also be investigated very carefully.   This should greatly influence the choice of workload measurement tool, but I fear it has been rather neglected in both the research and the applied settings. Here I review the state of the art with a view to identifying research that is needed in this domain of psychometrics – the intersection of sports science/medicine and biostatistics/epidemiology. I begin with the theory before I discuss how that can be relevant in sports science – specifically in injury—workload research.   1.3.1 Traditional measurement property views Traditional sport science/medicine research referenced the clinimetric literature to describe measures’ reliability, validity, and responsiveness [116]. Reliability quantifies a measure’s ‘repeatability’ or stability. Validity broadly asks whether something measures what it is supposed to measure but may also be divided into ‘types of validity’. Krosshaug and Verhagen [117] describe ‘face validity’ (whether a test appears to measure something), ‘content validity’ (determined by a team of experts who decide whether a measures captures what it intends to measure), ‘construct validity’ (whether a measure corresponds to a theoretical underpinning), and ‘criterion validity’ (correlation of a measure with a ‘gold standard’) [117].  20   Criterion validity can be further subdivided into concurrent validity – correlation with a gold standard at the same time – and predictive validity – correlation with a gold standard at a future point in time [117]. Responsiveness describes whether a given measure captures changes in what it is trying to measure over time [118]. For example, a fatigue measure should change following a heavy training or competition workload [119,120].  1.3.2 Unified validity theory Unified validity theory greatly influenced psychometrics in the last half century, but to my knowledge has not been introduced into sport medicine and science. Grounded in the work of Meehl, Cronbach, and others, US psychologist Samuel Messick primarily drove a unified view of validity theory, where construct validity takes centre stage [121–123]. Replacing ‘types of validity’, this unified validity theory proposed by Messick is defined in the Standards of Psychological Testing as follows:   “Validity is an integrated evaluative judgment of the degree to which empirical evidence and theoretical rationales support the adequacy and appropriateness of interpretations and actions based on test scores or other modes of assessment.” [122].  Unpacking this definition reveals the primary considerations, which differ from other, traditional validity views.  21 Validity is about inferences, not about measures Theoretically, we could measure the circumference around someone’s head just above the ears. This circumference provides information. We could use this information to infer what someone’s hat size may be, or we could make a conclusion about someone’s intelligence [118]. The inference in the first case is much more appropriate – more valid – than the latter. In this way, validity is less about the measurement itself, and more about the inference one intends to make from the measurement. There are multiple sources of validity evidence If validity is primarily concerned about whether a measure is appropriate, meaningful, and useful, it follows that people may ask to what degree this is the case. We might ask how the measure relates to other variables that are related, or unrelated, to our primary variable of interest. Do subject matter experts believe that our measure of interest is assessing what we think we are measuring? Does our measure differ between groups where we expect there to be differences? Is the internal structure of the measure related to the central idea we are hoping to assess? Other validity frameworks refer to each of these things as ‘types’ of validity or reliability, while unified validity theory articulates that each of these inquiries provide a different source of evidence to ascertain whether certain inferences are appropriate and meaningful given a certain measurement [124]. Validity is an integrative evaluative judgement An important concept, often overlooked by undergraduate students is that a measure cannot be ‘valid’ or ‘invalid’ [121,124,125]. Rather, whether a specific measure can be used for a specific 22  purpose is likened to a continuum that ranges from completely useless to extremely appropriate. To evaluate where an inference from a given measure sits, researchers must evaluate the strength and breadth of evidence surrounding a given measure [125]. How does the measure relate to other variables? Do changes on the measurement reflect changes in the gold standard measurement? Do experts agree the measure reflects the underlying construct it is designed to measure? What is the measurement really measuring, or what are participants thinking when they complete the measure? A unified approach identifies that multiple sources of evidence should be integrated when evaluating the reliability of certain measurement-informed inferences. How unified validity theory may contribute to workload—injury research Unified validity theory has the potential to inform validation work across sport sciences, including workload—injury research. In published manuscripts and in conference presentations, it is common to hear that a ‘validated measure’ was used. These statements often refer to a few papers where one or two sources of validity evidence was examined. By incorporating unified validity theory into sport sciences, researchers may consider the degree of confidence they have in their specific inferences, instead of determining a binary (valid/invalid) outcome for their measurements. When choosing a workload measure (internal or external), unified validity theory provides a lens for researchers to consider all the relevant sources of evidence and decide how appropriate workload measures and derived variables are for their specific sporting environment.     23  1.4 Methodological and statistical limitations in athlete monitoring and injury prevention research  Prospective cohort studies are the most frequent study design within the workload—injury literature [74,76]. Teams with the resources and budget to track workload longitudinally are often in elite sport environments, where performance is paramount and randomised controlled trials are not feasible. Therefore, most datasets capable of measuring workload and injury occurrences consist of longitudinal, observational data collected on several athletes on a regular (daily/weekly) basis. Considering the recursive nature of injury risk, varying across each session, it is important the data are collected frequent enough to capture these fluctuations, and statistical approaches are employed that can effectively analyse these changes. Professor Linda Collins described this principle in her threefold model of longitudinal data – where the theoretical model (aetiology), data collection frequency (temporal design), and statistical model should be as closely aligned as possible [126].  Longitudinal data present several challenges that researchers must account for when choosing an analytical approach. They must differentiate between- and within-subject effects [127], correlated errors of measurement (i.e. dependency caused by repeated measures), missing data, and potentially uneven time intervals [128]. Although there have been great advances in the area of longitudinal data analysis [128–130], the challenges above mean that certain statistical approaches are more appropriate. Although authors have systematically reviewed the workload—injury research and quantitatively synthesised the evidence, to date none have evaluated the workload--sports injury literature to assess the quality of statistical analyses.    24  1.5 Integrated athlete monitoring and injury research and practice Athletes’ training and competition workloads are strongly associated with injury risk [65,74,76,77]. How this fact impacts applied research and practice depends on (1) aetiological, (2) methodological, and (3) statistical considerations, and aligning these three areas.   Conceptually, existing aetiological models have not explicitly explained how workloads relate to injury risk, nor how they interact with other identified risk factors to culminate in sport injury occurrence. As aetiological understanding is among the first steps of injury prevention, researchers and practitioners should understand the mechanisms by which workloads contribute to sports injury.  Methodologically, applied researchers should consider what measures to collect in their specific sport settings. This decision requires an integrated, evaluative judgement of which measure is most appropriate for their purposes (in this case injury risk), considering all the available evidence. If they consider that daily workloads influence subsequent injury risk, researchers must also collect data frequently enough to capture these workloads and fluctuating risk.   Statistically, sports medicine researchers may only explore the workload—injury association adequately when they employ statistical analyses that address the challenges inherent with intensive longitudinal data.   Workload monitoring has taken center-stage in sports science/medicine research practice. A growing body of literature indicates that workloads are strongly associated with injury. However, 25  as technological advances allow more workload data to be collected than ever before, there is a need for aetiological and methodological frameworks to guide future workload—injury investigations.   26  Chapter 2: Aims of the Dissertation  My objective in this dissertation was to investigate a conceptual and methodological framework that explains how training and competition workloads are associated with sports injury risk.   My specific aims were,  1) To critically interrogate current sport injury aetiological models and ask why workload was not included in any of them. This led to my suggestion to update the workload—injury aetiology model to include training loads (Chapter 3). 2) To describe how workload data and risk factors (internal and external) may mediate or moderate the workload—injury association (Chapter 4). 3) To methodologically review the statistical techniques that authors have used to examine workloads and sports injuries using longitudinal data using Linda Collins’ three-part framework. (Chapter 5). This chapter includes a framework of best practice for longitudinal data analysis broadly and for scientists in the workload—injury field specifically.   4) Using the framework outlined in Chapter 5 and my subsequent recommendations for future investigations, to use multilevel modeling of intensive longitudinal data to examine the effect of player unavailability on players’ physical match workloads (Chapter 6). 5) To demonstrate how a statistical concern (mathematical coupling in the acute:chronic workload ratio) may inform workload—injury research and applied practice (Chapter 7). 6) To examine pre-season participation and in-season injury risk, accounting for in-season workloads using multi-level modelling (Chapter 8).   27  These 6 aims can be considered as 3 complementary parts, each addressing one component of my thesis title:  Part 1 (Chapters 3-4). Conceptual considerations of the workload—injury association,  Part 2 (Chapters 5-6). Methodological considerations of the workload—injury association, Part 3 (Chapters 7-8). Applied considerations of the workload—injury association.  28  Chapter 3: How do training and competition workloads relate to injury? The workload—injury aetiology model   3.1 Preamble Workloads had increasingly been associated with injury risk in the literature over the last decade, as described in Chapter 1. However, aetiological models to date had not included workloads explicitly in their frameworks, and could not adequately explain the workload—injury association. This discrepancy led to the focus of this chapter – an extended sports injury aetiology framework: the workload—injury aetiology model.   The rest of this chapter is a reproduction of text and figures from the following article [131]:  Windt J, Gabbett TJ. “How Do Training and Competition Workloads Relate to Injury? The Workload—Injury Aetiology Model.” British Journal of Sports Medicine, July 14, 2016, bjsports-2016-096040.  3.2 Introduction Athletic injuries are common in a variety of sports [132,133], compromising performance [52,53,134], posing a financial burden to organisations, [54] and potentially causing long-term health consequences [135]. The causes for these injuries are numerous, highlighted by a number of multifactorial injury etiology models [4,36,69]. However, regardless of the interplay of risk factors or inciting biomechanical event, every athletic injury is sustained while athletes are exposed to the demands of the sport, either through competition, or through practice workloads 29  that are applied to athletes with the goal of inducing positive physiological changes and maximising performance.   While adequate workloads are necessary for fitness and performance improvements through adaptation and skill acquisition, high workloads and especially ‘spikes’ in loads are strongly associated with injuries [52,113,115,136,137]. Although this association has been demonstrated in soccer [75], Australian Football [138–140], basketball [141], rugby union [52], rugby league [113,142,143] and cricket players [136,144,145], few have explained the underlying mechanisms through which workloads relate to injuries. Moreover, existing injury etiology models cannot account for the relationship between workloads and injury risk. The purpose of this paper is to present an updated etiology model that accounts for the effects of workloads on athletic injuries. After reviewing previous injury etiology models, we discuss the known association of workloads and injuries, and present an updated model that explains a threefold contribution of workloads to injury risk. Finally, we discuss the research and practice implications of the updated model.   3.3 Why do athletes get hurt? Injury etiology models  The multifactorial nature of athletic injuries is well known [4,69,71]. Single risk factors fail to adequately predict injuries, and the limitations of univariate analyses of risk factors have been discussed [69]. Instead, multivariate techniques are recommended, which enable researchers to examine multiple risk factors which come together to cause an injury [69,71]. The first of these multifactorial models was proposed by Meeuwisse in 1994 [69]. Meeuwisse built upon chronic disease models to propose an epidemiologic, multifactorial approach to understanding injury causation and investigating athletic injuries (Figure 3.1). Within this multifactorial model, athletes 30  have intrinsic (internal) risk factors that inherently predispose them to injury. Some of these factors (e.g. age) are non-modifiable, while others (e.g. flexibility) are modifiable. These predisposed athletes are then exposed to extrinsic (external) risk factors, such as the playing surface, protective equipment, or opponent behaviour, which make them susceptible to an injury. Finally, a certain inciting event occurs wherein the biomechanical stress of the event exceeds the tolerance of the athlete’s tissues, and injury ensues. Lastly, Meeuwisse also acknowledged that risk factors are interrelated, and in a subsequent article addressed the importance of interpreting both interaction and confounding effects in injury etiology research [70].31    Figure 3.1: Meeuwisse’s first multifactorial injury etiology model proposed in 1994.  With permission, from [69]   32  In two subsequent papers published in the British Journal of Sports Medicine [36,71], Bahr and colleagues built upon this first multifactorial model, highlighting and focussing on certain aspects (Figure 3.2). First, they expanded the description of both internal and external risk factors and the subsequent methodological implications [71]. Second, they proposed a more thorough description of the biomechanical factors contributing to the inciting event [71], based on the comprehensive biomechanical injury model of McIntosh [146].  33   Figure 3.2 Bahr & Krosshaug’s 2005 comprehensive etiology model.  BMD = bone mineral density, ROM = range of motion. With permission, from [36]. 34  Most recently, Meeuwisse and colleagues proposed a modification to these etiology models in the form of a “dynamic, recursive injury etiology model” (Figure 3.3) [4]. This dynamic model of risk and causation acknowledges that previous linear models do not account for the dynamic and non-linear nature of athletes’ injury risk.  Moreover, sport participation may or may not result in an injury occurrence. In cases where participation doesn’t result in injury, it actually modifies certain risk factors through physiological adaptation to the training stress. On the other hand, if an athlete does experience an injury, there is also acknowledgement that an athlete may either recover and return to play with a modified injury risk, or never recover, resulting in complete removal from participation. 35   Figure 3.3: Meeuwisse and colleagues’ 2007 dynamic, recursive sports injury etiology model.  With permission, from [4]. 36  This continued evolution in understanding injury etiology helps to inform injury prevention efforts. It provides a framework in which investigation scope can be defined, indicates appropriate statistical analyses, highlights important variables to consider, and in the case of the dynamic model, reminds researchers that participation not only predisposes to injury, but modifies subsequent risk. Further, the consistent modification of the model also highlights the need for further research, and willingness to amend the model as more knowledge is accumulated [147].  Notably, workloads are not included as an internal risk factor nor an external risk factor in existing etiology models. In fact, although workloads are strongly associated with injury [52,113,115,136], their role in the etiology of athletic injuries is uncertain, given that they are not explicitly included anywhere within the previous models.   3.4 Workloads and injuries – what do we know?    3.4.1 Defining and quantifying workloads  Workloads, as defined in Gabbett and colleagues’ systematic review on workloads, performance and injury, are “the cumulative amount of stress placed on an individual from multiple training sessions and games over a period of time” [75]. As Smith explains, they are “a combination of … [training and competition] intensity, duration and frequency” [78]. Essentially, workloads are a quantification of the demands imposed on an athlete during one or more matches or training bouts. Ideally, training is prescribed in such a way that the athletes’ homeostasis is disrupted, and optimal adaptation occurs during recovery. This fine balance seeks to avoid, on the one hand – insufficient workloads that fail to induce adaptation or result in detraining, and on the other hand – excessive loads that induce maladaptation or overtraining [78,96,109]. However, optimally prescribing these 37  loads presents a number of challenges, the most notable of which is choosing an appropriate workload measure.   Workloads can be measured as either external or internal loads. External loads quantify the amount of work performed by the athlete (e.g. distance covered, balls thrown, etc.), while internal loads measure the “relative physiological and psychological stress imposed” on the athlete [96]. A given external load will elicit different internal responses in each athlete, based on the characteristics of how the external load is applied and the athlete’s individual characteristics (e.g. genetics, fitness level, training background, etc.) [97]. Numerous measures have been developed and proposed for internal and external loads from which practitioners can decide (Table 3.1) [96,101].  However, each of these measures aims to quantify the workload completed by the athlete and/or their response to that work. Careful consideration must be given to which type (internal or external) and specific measure(s) of load are most appropriate, given the sport context, goals of load monitoring, logistical and financial constraints, and the psychometric properties (validity/reliability) of the specific measure.    38  Table 3.1: Measures of external and internal load External Load Internal Load - Frequency (/day, /week, /year) - Time (minutes, seconds) - Accelerometer loads - Distance covered  - High speed distance covered - Jumps completed - Power output, speed, acceleration - Time-motion analysis - Neuromuscular function - Weight lifted - Throws/Pitches/Bowls performed - Session-Rating of Perceived Exertion (RPE*minutes) - HR:RPE Ratio - TRIMP (HR-based training impulse) - Blood Lactate - Lactate:RPE Ratio - Recovery/stress/wellbeing questionnaires (e.g. Recovery Stress Questionnaire for Athletes - REST-Q; Daily Analyses of Life Demands of Athletes – DALDA, Profile of Mood States - POMS)  Various examples for quantifying both external and internal workload. Examples taken from [96,101,111]  3.5 Monitoring workloads in relation to performance, periodisation, and overtraining As Borresen and Lambert describe in their review of workloads, physiological responses and performance, “optimising training first involves quantifying what the athlete is currently doing” [101]. Traditionally, this quantification of load has allowed, 1) performance to be predicted, 2) training to be planned, periodised, and tracked, and 3) training readiness and stress to be monitored.  1 – Predicting performance. In the mid-1970s, Banister and colleagues proposed the first systems model of performance, characterising the athlete as a system with input (workload) and subsequent output (performance) [105]. Their landmark model proposed fitness as a positive impulse of training, and fatigue as a negative impulse, acknowledging the dual forces of stress/breakdown caused by a training bout and the potential for positive adaptation with sufficient recovery. Within 39  this model, both impulses decline exponentially once a session ends and recovery begins, but fatigue decays at a much more rapid rate than fitness.    These basic principles underpin performance strategies like tapering, where a 3% (ranging 0.5-6.0%) improvement in performance may be attained through altering training volume and intensity leading into peak competitions [103,104,148]. While attractive in theory and implemented widely in practice, Banister’s model and similar performance prediction models [106,107] have some limitations in their ability to achieve high levels of precision [149]. These limitations may be partly attributable to individually variable load—adaptation responses to set training regimes [150,151]. It is thus important that these basic principles should be adapted to individual athletes, with the applied load and individual response monitored where possible [82].   2 – Planning, periodising, and implementing training plans. Coaches use the inherent principles of workloads and adaptation to design periodised training plans, with the goal of eliciting optimal adaptation [86,152]. However, workloads must be monitored if coaches are to ensure that their training plan is implemented as intended, as the discrepancy between coaches’ and athletes’ perceptions of training has been demonstrated [102]. Further, workload monitoring is necessary to make and implement refined alterations to training loads. For example, these precise refinements are implemented during tapering phases where training volume may be exponentially reduced while intensity remains high, with the goal of minimising cumulative fatigue and peaking for competition [103,104,148].  40   3 – Monitoring fatigue to prevent overtraining. Monitoring workloads over time can also be utilised to assess athletes’ stress levels, fatigue, mood, and readiness to train [96]. Both physiological measures (e.g. heart rate variability, heart rate recovery, hormone levels, catecholamine levels) and psychological measures (e.g. REST-Q, POMS, DALDA) have been recommended to monitor athletes’ internal workloads and responses [109,153–155]. In monitoring workloads longitudinally, large individual deviations from normal responses, and discrepancies between internal and external load measures can be used to assess athletes’ responses to training [96]. For example, lower internal loads with a standard external load may indicate fitness improvements, while increased internal loads with the same external load may indicate a state of fatigue [96]  In all of these traditional uses, training loads are imposed on athletes with the goal of maximising performance, acknowledging that while training is necessary to induce the “reward” of positive adaptations, it carries the known “risk” of imposing stress upon the athlete. Thus, every time an athlete trains, they are exposed to the fatiguing effects of training, as well as risks of potential maladaptation through overtraining. However, excluded in these traditional uses and this aforementioned risk – reward balance is the reality that every training and competition load carries the potential for athletic injury [156]. In fact, every athletic injury is sustained while an athlete performs some sort of workload. This begs the question: how do workloads relate to injury occurrence? Is it a simple matter of increased loads leading to increased injury, or is there more at play? Recently, researchers have begun providing some answers in numerous sporting contexts and with various load measures – teasing out this association.   41  3.5.1 Total workloads and injuries  With each training and competition bout, athletes are exposed to the risk of sustaining an injury, so it may seem intuitive that increased loads should result in increased injury levels. Traditionally, workload-injury investigations focused on this relationship between absolute workloads and injury [141,142,157–159]. To highlight a few examples, rugby players running more than 9 m of very high speed running were 2.7 times more likely to experience a non-contact injury than those below that threshold [160].  Australian Football (AFL) players with higher total distance (odds ratio 5.5) and sprint distance (odds ratio 3.7) accumulated over 3 weeks were at a greater likelihood of injury [138]. Furthermore, higher pitching loads in baseball [161,162], and bowling loads in cricket have also been associated with increased injury risk [144].   Gabbett developed an injury prediction model using session-rating of perceived exertion (s-RPE) as a marker of internal load among elite rugby league players over two years [115]. During the subsequent 2 years, the model was used to predict injuries, showing that players who exceeded the weekly workload threshold as determined by the model were 70 times more likely to test positive for non-contact, soft-tissue injuries, while players who did not exceed the threshold were injured 1/10 as often [115]. Collectively, these data indicate that there is an increase in injury risk with absolute workload increases.   This higher workload - higher injury relationship may create the impression that workloads should be kept low in order to minimise injury risk. While a reduction in training load may be appropriate in certain instances [158], two important points should be stated. First, adequate workloads are necessary to induce beneficial physiological adaptations such as high aerobic capacity, optimal 42  body composition, strength, and repeat-sprint ability [163–165], which are required for high performance, and many of which are associated with decreased injury risk [160,166,167]. Thus, workloads that are too low may not only decrease performance, but may result in lower levels of fitness and preparedness, subsequently increasing injury risk. For example, a ‘dual threshold’ has been shown in cricket fast bowlers, where not only high workloads and short recovery periods associated with increased injury risk, but a low total number of deliveries (<123, relative risk =  1.4) and long recovery periods (>5 days, relative risk = 1.8) also increased injury risk [144].  Second, it may not solely be the total workload that is applied that contributes to injury risk, but the way in which it is applied. Indeed, some investigations show that total workloads are not always associated with increased injury risk, while the rate of change in these workloads over time is a stronger predictor [112,113,136].    3.5.2 Acute:Chronic workload ratio and injuries Specifically, load-injury investigations have recently examined the relationship between acute (1 week) and chronic (4 week) workloads, termed the acute:chronic workload ratio, with injury risk [52,113,136]. In the first study of acute:chronic workload ratios, Hulin and colleagues showed that while internal and external loads of cricket fast bowlers in isolation were not significantly related to injury risk, acute:chronic workload ratios of >1.5 increased injury risk by 2-4 times in the subsequent 7 days [136].   High acute:chronic ratios, simply described as training ‘spikes’, similarly increase injury risk in rugby league players [113,168] and professional soccer players [169], with utility for use in 43  predicting injury (Figure 3.4) as well as subsequent injury during athletes’ return to play [170] (Figure 3.5).  In addition to understanding that both total workloads and changes in load over time are related to injury risk, researchers have begun combining these two variables to determine the optimal outcome for both injuries and performance.     Figure 3.4: Injury likelihood associated with various weekly changes in workloads.  Unpublished data collected from professional rugby league players over three pre-season preparation periods. Loads were measured using session-rating of perceived exertion, and a total of 148 injuries were sustained. Reused, with permission, from Gabbett [112].  44      Figure 3.5: Finding the ‘sweet spot’ for acute:chronic workload ratios in return-to-play protocols. Data redrawn from 3 separate sports (cricket, rugby league and Australian football) demonstrating injury likelihoods associated with various acute:chronic workload ratios. With permission [112].   3.5.3 Workload – injury paradox In contrast to the idea that higher workloads contribute to higher injury incidence, high workloads may contribute to well-developed physical qualities, thereby reducing injury risk [112,171,172]. The training--injury prevention paradox model has recently been defined and reviewed by Gabbett – describing this chronic load effect [112]. This ‘paradox’ of high chronic workloads is their 45  potential preventative effect, as long as the acute:chronic workload ratio is kept in a moderate range (i.e. training spikes are avoided).   Hulin and colleagues found that elite rugby players who had very high acute:chronic workload ratios as well as high chronic workloads had the largest risk of injury (28.6%). However, they also demonstrated that as long as acute:chronic workload ratios were kept within a moderate zone (0.85-1.35), high chronic workloads were associated with the lowest risks of injuries, other than very low (< 2 SD) acute:chronic ratios (Table 3.2) [113]. In a subsequent rugby league study, the odds of sustaining an injury during the competitive season were decreased by 17% for every 10 pre-season sessions that players completed, controlling for in-season workloads [173].     46  Table 3.2: Relationships between injury risk and various acute:chronic workloads combined with high and low chronic workloads Acute:chronic workload ratio Combined with low chronic workload Combined with high chronic workload Very-low 0.0 ± 0.0 0.0 ± 0.0 Low 7.8 ± 4.1 9.6 ± 4.1 Moderate-low 10.0 ± 2.5 7.5 ± 2.2 Moderate 9.3 ± 2.6 6.2 ± 2.2 Moderate-high 11.0 ± 4.9 7.1 ± 4.0 High 5.9 ± 7.3 12.0 ± 10.7 Very-high 18.2 ± 14.9 28.6 ± 18.1 Risk of injury (±90 CI) associated with 2-week average acute:chronic workload ratios combined with low (<16 095m) chronic workload, and high (>16 095m) chronic workload in elite rugby league players. With permission, from Hulin et al. [113]  As extensively discussed in Gabbett’s recent review [112], these findings support the notion that workloads should be high enough to produce the beneficial physiological adaptations which may protect against injuries. However, these beneficial effects can be negated if athletes are exposed to large workload spikes, which are likely responsible for a large proportion of non-contact soft tissue injuries [112]. Therefore, the collective findings of workload-injury data indicate that while increased loads are somewhat associated with increased injury rates, optimal load management to minimise injuries entails the accumulation of high chronic workloads, while minimising spikes by maintaining week-to-week changes within approximately 10% [112]. What remains to be explicated are the mechanisms by which these findings link to our current knowledge of injury etiology.   47  3.6 An updated injury etiology model – incorporating the effects of workloads  Previous injury etiology models neither included workloads within the model, nor explained the strong association of loads with injuries. We have learnt a great deal about the association of workloads with injuries and it is clear that the application of training or competition loads are not inciting events that directly cause injury. An athlete may have a large spike in training and subsequently rupture his or her anterior cruciate ligament, but this injury still occurs via some inciting event (e.g. a dynamic valgus collapse). Overuse injuries, or ‘training load error injuries’[174], may be strongly workload-related, but still have an inciting event of cumulative tissue overload, even though a specific event occurrence may not be identifiable [16,19].   Further, training loads are neither a characteristic of the athlete (internal risk factor), nor an aspect of the environment in which the athlete participates (external risk factor). Rather, training and competition loads are better understood as the ‘vehicle’ in which athletes are exposed to external risk factors and potential inciting events. With this understanding, injuries are not directly caused by workloads. Instead, training and competition loads contribute to injury risk through exposing athletes to potentially injurious situations, as well as through their positive and negative effects on numerous modifiable internal risk factors. Therefore, we propose an updated injury etiology model that explicitly incorporates workloads within the causal chain, and outlines its known effects (See Figure 3.6).   48    Figure 3.6: The workload – injury etiology model. 49  Like Meeuwisse’s 2007 dynamic recursive model [4], our model depicts the ever-changing nature of an athlete’s injury risk, as well as the possibility that an athlete may, or may not, experience an injury after being exposed to training or competition loads. We have retained the role of internal and external risk factors in line with previous model iterations [36,71]. However, we have divided internal risk factors into modifiable and non-modifiable factors to differentiate those which can change through adaptations from those that are stable. Further, we have extended previous models by including the ‘Application of Workload' as the primary process whereby an athlete is exposed to various external risk factors, and the potential for inciting events, thereby moving from a ‘predisposed athlete’ to a ‘susceptible athlete’,   Whereas Meeuwisse’s 2007 model specified that an athlete who engages in repeat participation may experience adaptation/maladaptation from training and thereby be at an altered risk [4], we explicitly outline that athlete adaptation comes as the result of each applied workload. As in Banister’s original performance prediction models, we highlight that are both positive adaptations from training (i.e. fitness), and temporary negative effects of training (i.e. fatigue). Whereas Banister and colleagues focused on these adaptations from a performance perspective, we contend they function similarly in dynamically effect injury risk. These adaptation effects alter an athlete’s modifiable internal risk factors so that when they engage in a subsequent training or competition bout, their injury predisposition will be altered.   Lastly, we note that when an athlete experiences an injury, they must engage in the rehabilitation and return to play process. In this scenario, an athlete’s subsequent participation occurs with a modified injury predisposition due to the presence of an index injury [58]. However, in 50  rehabilitating and preparing to return-to-play, these athlete’s rely on the application of workloads through the same process to restore resilience in the injured tissue and prepare them for the demands of training and competition [170].  3.6.1 The triple role of workloads – ‘exposure’, ‘fitness’, and ‘fatigue’ Within the updated model, a predisposed athlete becomes susceptible to injury when they are exposed to a training or competition load. This workload exposes them to external risk factors and potential inciting events for the duration of the bout. Further, workloads modify subsequent injury risk. This risk modification occurs through both positive and negative adaptations, dictated by the total and relative workload applied. Therefore, workloads affect injury etiology in 3 primary ways.  1 - Exposure – training and competition loads are the medium exposing athletes to external risk factors potential inciting events.  2 - Fitness - positive adaptations associated with training, which improve modifiable internal risk factors, such as aerobic capacity, skill level, or body composition.  3 - Fatigue - negative adaptations associated with training, temporarily causing decreased capacity in modifiable internal risk factors, such as tissue resilience or neuromuscular control.  3.6.2 How do workloads influence injury risk? Implications of the updated model According to this model, workloads are most likely associate with injuries under three main conditions:    51  1 – High workloads increase exposure, thereby increasing injury risk.   Since training and competition loads are the medium exposing athletes to potential inciting events and external risk factors, there will always be, to some degree, a higher load – higher injury relationship. No load, and thus no exposure, cannot result in athletic injury, regardless of the internal and external risk factors at play. This aspect of the model explains a great deal of the literature which has found increased injury risks with increased workloads [141,142,157–159].   2 - Workloads which induce high levels of negative changes to modifiable internal risk factors (i.e. ‘fatigue’) increase injury risk.  A given external workload is a poor predictor of fatigue, since individuals vary widely in their internal response to set loads. A better marker of how workload effects fatigue may be the acute:chronic workload ratio, wherein a high ratio indicates that an athlete has been placed under a load that is substantially higher than their recent training has prepared them for. Physiological and psychological measures of internal load can also indicate an athlete’s fatigue levels. In this case, negative changes may occur to a number of modifiable internal risk factors, such as compromised neuromuscular control or reduced tissue resilience, which increase subsequent injury predisposition.  3 – Workloads that maximise positive adaptations while minimising fatiguing effects will help make athletes more robust to injury.  On the other hand, positive adaptations to modifiable internal risk factors help to explain the training-injury prevention paradox. As long as the acute:chronic workload ratio is kept moderate, implicating lower fatiguing effects, higher chronic workloads may be preventative against injury 52  due to the positive adaptations of training [112]. From a biomechanical perspective, Bahr explains “improved fitness may protect the tissue against injury through the effects of training on its material properties” [36]. Together, acknowledging the recursive nature of workload adaptations with both positive and negative adaptations helps to frame why the acute:chronic workload ratio is strongly associated with injury, and why high chronic workloads may be protective against injury [112,113,136].   3.6.3 Model limitations As with any model, our proposed injury etiology model is incomplete. A set workload may induce various levels of positive and negative adaptations that are based upon additional athlete stressors (e.g. travel, lack of sleep), which are not explicitly incorporated into the model. Some known risk factors are not explicitly listed (e.g. genotype, psychological state, muscle fibre type, etc.) and some that are yet to be identified. As just one example, workloads may effect certain psychological variables (e.g. commitment to training, self-blame, perceived stress/fatigue, etc.), which could subsequently alter match/training behaviour and change injury risk [155,175]. Moreover, the interactions/confounding relationships between various risk factors are not detailed. Some of these, such as the interaction between previous injury and neuromuscular control are known [36], while others are yet to be explained. However, even just including the known confounding and interacting relationships would make the model nearly illegible due to the number of interrelating connections.   It should be noted that workloads are unlikely to contribute equally to all injuries, on the basis of different tissue characteristics and mechanisms of injury [176]. Different tissues may respond 53  differently to loads, exemplified in cricket fast bowlers, where high medium term (3 month workloads) are protective against tendon injuries, but a risk factor for bone stress injuries [176].  Further, workloads may contribute less to contact injuries than non-contact injuries. For example, workloads relate to an ice hockey player’s concussion following an illegal body check primarily by providing the exposure to certain external risk factors (e.g. opponent behaviour, a compromised helmet). Conversely, a non-contact injury such as a soccer player’s ACL rupture while changing direction is likely attributable to a greater number of internal risk factors. In this instance, workloads contribute again in this scenario through exposure, but loads in the weeks preceding injury may also have modified certain internal risk factors – tissue resilience, neuromuscular control, etc. – which further predisposed the athlete to injury. Therefore, a number of investigations relating load and injury have focused on non-contact soft-tissue injuries, which are considered ‘preventable’, more sensitive to load changes, and more attributable to internal risk factors [112,115].   Similarly, workload may contribute differently to overuse vs. acute injuries. Whereas workloads may modify athletes’ internal risk such that they are predisposed to an inciting injury event, overuse injuries may occur without a single identifiable injury event. However, the suggested model indicates that this type of cumulative tissue overload could occur with exceedingly high workloads, or spikes in workloads, that result in high levels of fatigue (negative adaptations) without adequate recovery for positive adaptations.   54  3.7 Where to from here – research implications   The dynamic nature of the current model presents similar challenges to research design and statistical analysis as Meeuwisse’s dynamic recursive model [4]. From a design perspective, although non-modifiable risk factors can be measured once (since they will not change), modifiable risk factors should be measured repeatedly to account for changes over time. Repeated measurement are even more important in the case of training and competition loads, since quantifying both total loads and acute:chronic workload ratios can only be done if workloads are consistently monitored over time. Although repeated measurement presents logistical and financial challenges, repetition is necessary to quantify the dynamic nature of modifiable factors, workloads, their interactions, and athletes’ subsequent injury risk.   From a statistical analysis perspective, repeated measurement also implicates certain statistical analyses. Simple logistic regression models which assume the same exposure (training and competition load) across individuals may not be appropriate [71], given the large variability in workload measures across individual players. Although Cox proportional hazards regression (time-to-injury event) and Poisson regression (rate of injury per 1000 exposures or per 1000 hours) have been presented as potential models to account for time exposure [4], they often fail to account for the intensity of training which is captured in various load measurements. Though not frequently used, multilevel (i.e. mixed) modelling may allow for these analyses, since they can account for correlated outcomes (repeated measures among players), and include random effects to predict individual athletes’ risks [127,177,178]. The frailty model which also allows random effects for players and the ability to control for the dependencies of recurrent injuries and repeated measures also shows promise [38,44]. 55   3.8 Summary Training and competition workloads have been an obvious omission from previous injury etiology models. Workloads are neither an internal nor external risk factor for injury, but are the ‘vehicle’ by which athletes either compete in, or train for their respective sports. We propose that our updated injury etiology helps provide a conceptual framework for ‘why’ workloads are strongly associated with injuries. According to this model, increased workloads increase injury risk through exposure to external risk factors and potentially inciting events. However, they also continually modify injury predisposition by inducing physiological adaptations, some positive, and others negative. Therefore, when an athlete experiences a high acute:chronic workload ratio, they likely experience larger degrees of negative maladaptation, modifying a host of internal risk factors and increasing their predisposition to injury in subsequent bouts. Conversely, moderate acute:chronic workloads and the accumulation of high chronic workloads may maximise the positive physiological adaptations to training and thereby reduce injury risk, as suggested by the training load-injury prevention paradox model.  We note that the confounding and interacting effects of workloads and many risk factors are still unknown. However, longitudinal investigations with repeated measures of workloads and modifiable risk factors may begin to unravel these relationships and provide further insight into the dynamic nature of injury etiology.   56  Chapter 4: Why do workload spikes cause injuries, and which athletes are at higher risk? Mediators and moderators in workload–injury investigations  4.1 Preamble In chapter 3, I introduced a new athletic injury aetiology model that explicitly included training and competition workloads (Figure 3.6). The model considered traditional internal and external risk factors, but also acknowledged how workloads contribute to the recursive nature of injury risk – through positive physiological adaptations (i.e. fitness) and transient negative effects (i.e. fatigue). Soon after the first multifactorial model of athletic injury, Meeuwisse discussed how different risk factors relate to each other, and how to differentiate the concepts of interaction and confounding [70]. Extending this aetiological thinking to the workload—injury aetiology model, I expanded how traditional risk factors may interact with workload, specifically through mediation or moderation.   The rest of this chapter is a reproduction of text and figures from the following article [179]:  Windt J, Zumbo BD, Sporer BC, MacDonald K, Gabbett TJ. “Why Do Workload Spikes Cause Injuries, and Which Athletes Are at Higher Risk? Mediators and Moderators in Workload–Injury Investigations.” Br J Sports Med, March 4, 2017, bjsports-2016-097255.  4.2 Introduction  Spikes in training and competition workloads, especially in undertrained athletes, increase injury risk [112]. However, just as attributing athletic injuries to single risk factors is an oversimplification of the injury process [131,180], interpreting this workload—injury relationship 57  should not be done in isolation. We must further unpack how (i.e. through which mechanisms) workload spikes might result in injury, and what characteristics make athletes more robust or more susceptible to injury at any given workload. In other words, which factors mediate the workload—injury relationship, and which moderate the relationship.   4.3 Domino or dimmer? Differentiating ‘mediators’ and ‘moderators’  Like dominoes being knocked over, mediators can be viewed as the intermediary steps that explain the association between an observed variable and an outcome [181]. In this context, mediating variables help to explain ‘why changes in workloads might cause injuries’? For example, it is known that rugby league players exposed to spikes in running workloads, indicated by a high acute:chronic workload ratio, are at an increased risk for non-contact injuries [182]. One potential explanation is that neuromuscular fatigue mediates this relationship, such that increased workloads cause higher levels of neuromuscular fatigue, subsequently precipitating injury (e.g. while performing a cutting manoeuvre) (Figure 4.1).    On the other hand, moderators can be viewed as ‘dimmer switches’, modifying the effect of a given variable on an outcome. They are also referred to as interactions, or effect modifiers [70].  In our context, moderator variables answer the question: ‘what characteristics make certain athletes more robust or more susceptible to injury at given workloads’? For example, in Gaelic football, high aerobic fitness protects against workload spikes [182]. In other words, aerobic fitness protectively moderates the workload effect by ‘dimming’ or reducing the risk of rapid workload increases (Figure 4.1).    58   Figure 4.1: Examples of how mediating and moderating variables explain the association between a workload spike and subsequent injury.  Top, neuromuscular fatigue is shown as a mediator (i.e. domino) of the effect of workload spikes on injury. In this example, a spike in workload causes an increase in neuromuscular fatigue, and an increase in neuromuscular fatigue leads to an injury. Bottom, aerobic fitness acts as a moderator (i.e. dimmer switch) of the relationship between workload spikes and injury. In this case, a given spike in workload will elicit a different injury risk dependent on an individual’s fitness level.  59  4.4 Not so simple, not so fast Admittedly, these previous examples are simple illustrations of how mediation and moderation are conceptualised. In reality, the aetiology of injury is complex, dynamic, multifactorial, and context-dependent [131,180]. Therefore, for certain injuries, the effect of workload on injury is most likely more appropriately viewed as moderated mediation [181]. Combining the two aforementioned examples, a spike in workload may produce increased levels of neuromuscular fatigue, but the strength of that relationship may be moderated by aerobic fitness. In complex systems language, this example may manifest in a ‘risk profile’ which includes workload, aerobic fitness and neuromuscular fatigue as interacting factors within a ‘web of determinants’ – related to an ‘emergent pattern’ of non-contact injuries [180].  Expounding the causal relationship of workloads and injuries and explaining more complex causal pathways presents both methodological and analytical hurdles. First, study design and data collection must be conducted in a way so that mediator and moderator variables are both considered and measured. To reduce the plausibility of alternative explanations, randomised experimental designs should be considered where possible [181]. The a-priori selection and measurement of potential confounding variables, which do not lie on the causal pathway, should be simultaneously conducted [70]. Incorporating these additional variables requires greater sample sizes (i.e. increased power) and a large number of injuries to demonstrate these mediation/moderation effects [181]. Finally, adequate analysis of mediation and moderation often requires both complex designs, and more complex analyses, including structural equation modelling and multilevel modelling, to name a few [181].   60  4.5 What rewards await at the finish line after jumping the hurdles?  1) Improved understanding, better prevention – Mediation and moderation, as well as their more complicated cases, will improve our understanding of causal mechanisms behind injuries, and subsequent preventative strategies. Moderating variables which are modifiable are key intervention points to ‘dim’ the risk associated with training and competition workloads.   2) Addressing insufficiencies – The inability of screening single risk factors to predict future injuries has been understandably challenged [183]. However, shifting our conceptualisation of these variables from injury predictors to variables that moderate the load—injury relationship provides a different motivation for screening and addressing any shortcomings. For example, while aerobic fitness may not predict injuries, it may increase athletes’ resilience to higher workloads. Therefore, when athletes perform poorly on pre-season fitness tests, practitioners may choose to lower the allowable workload ‘threshold’ for this athlete, while providing individualised attention to address the deficiency.    3) Training smarter and harder – Ultimately, attaining high chronic training loads without rapid spikes in the process is considered current best-practice [112]. However, understanding the characteristics that make certain athletes more robust may allow for more nuanced training load prescriptions. For athletes with a collection of characteristics which ‘dim’ workload-related injury risks (e.g. high aerobic fitness, no previous injuries), practitioners may consider prescribing higher training workloads (e.g. an acute:chronic workload ratio of 1.7) for performance purposes, given that their associated risk is lower than that of the average athlete.     61  Chapter 5: Getting the most out of intensive longitudinal data: A methodological review of workload—injury studies (Paper 3) 5.1 Preamble In the previous two chapters I described how athlete workloads and risk factors (internal and external) relate to injury risk. These aetiological models detailed that injury risk changes on a daily or per-session basis as workloads alter athletes’ individual risk profiles (either positively or negatively). Although these provided conceptual frameworks through which workloads relate to injury risk, the recursive, constantly changing nature of injury risk implied by these models makes analysing this association a challenge. Systematic reviews and consensus statements [74–76] and an International Olympic Committee Consensus Statement [65] have synthesised the existing workload—injury literature, but such reviews have not scrutinised  the methodologies employed by the authors.   As part of my critical literature review (Chapter 1), I identified challenges in analysing intensive longitudinal data, such as workload—injury analysis. Specifically, I identified generalised estimating equations were often used to account for the dependency created by repeated measurements in these studies despite multilevel (i.e. mixed) modelling approaches having several advantages. This chapter reports my detailed review of this topic. It is in press at BMJ Open (accepted September 4th, 2018) and the rest of this chapter is a reproduction of text and figures from that manuscript [184]:  Windt J, Ardern CL, Gabbett TJ, Khan KM, Cook CE, Sporer BC, Zumbo BD. “Getting the Most out of Intensive Longitudinal Data: A Methodological Review of Workload—Injury Studies.” BMJ Open. In Press. (Accepted September 4, 2018) 62  5.2 Introduction Intensive longitudinal data (ILD) are being collected more frequently in various research areas [130], catalysed by technological advancements that simplify data collection and analysis [185]. By collecting data repeatedly on the same participants, researchers are enabled to answer more detailed research questions, particularly regarding phenomena that change or fluctuate over time. However, arriving at these answers requires researchers to overcome the challenges of analysing ILD, which include: (1) the dependencies created by repeated measures, (2) missing/unbalanced data, (3) separating between- and within-person effects, (4) time-varying and time-invariant (stable) factors, and (5) specifying the role of time/temporality [186].  The field of exercise and sports medicine provides one specific example that can illustrate principles that apply to the use of intensive longitudinal data broadly. In the field of sports performance, technological advances mean that a plethora of physiological, psychological and physical data are conveniently available from athletes [96,187]. As one example, of 48 professional football clubs that responded to a survey on player monitoring, 100% reported collecting daily global positioning system (GPS) and heart rate (HR) data [98].   One research question that has gained a great deal of interest in the last decade is how athletes’ training and competition workloads relate to injury risk. Since athletes’ training and injury risk continually varies over time, many researchers have used prospective cohort studies to collect and analyse ILD to answer this question [76]. There is moderate evidence from systematic reviews and an International Olympic Committee (IOC) consensus statement suggesting a positive relationship between injury rates and high training workloads, increased risk of injury with low workloads, and 63  a pronounced increase in injury risk associated with rapid workload increases [65,74–76,188]. However, such systematic reviews do not consider the statistical approaches used in included studies [189]. Choosing the wrong statistical analysis or poorly implementing an otherwise correct one (e.g. violating statistical assumptions) can bias results and create false conclusions. Even a perfectly performed systematic review cannot compensate for poorly designed, or poorly analysed studies [190].     Longitudinal data analysis is most effective when the chosen statistical approach aligns with the frequency of data collection and with the theoretical model underpinning the research question (See Box 1) [126]. Therefore, we used this lens to evaluate the statistical models employed in prospective cohort studies using ILD to investigate the relation between athletic workloads and injury. We had 3 aims: (1) to summarise researchers’ data collection, methodological, statistical, and reporting practices [189,191]; (2) to evaluate the degree to which the adopted statistical analyses fit within Collins’ three-fold alignment [see Box 1]; and (3) to provide recommendations for future investigations in the field.   Box 1: Theoretical model – temporal design – statistical model.  In a landmark, highly-cited paper, Professor Linda Collins described how aligning the (1) theoretical model (subject matter theory), (2) temporal design (data collection strategy/timing), and (3) statistical model (analytical strategy) is crucial when analysing longitudinal data [126]. For example, if researchers (1) theorise that a given physiological variable fluctuates every hour, (2) data must be collected at least on an hourly basis. If researchers measure participants once a day, they will miss virtually all the hourly fluctuations that their theories predict. Once researchers 64  have collected their hourly data, they should (3) select a statistical strategy that enables them to examine the relationship between these fluctuations and the outcome of interest. As Collins noted, perfect alignment of these 3 components may not be possible, but it provides researchers a target, and readers a lens through which longitudinal research can be evaluated.   5.3 Methods 5.3.1 Article selection We systematically searched the literature (MEDLINE, CINAHL, SPORTDISCUS, PsychInfo, and EMBASE) (December 10, 2016) to identify systematic reviews and consensus statements that investigated the relationship between workloads and athletic injuries, with the aim of extracting all original articles included in these reviews that met our inclusion criteria. A summary of the systematic search and article selection process is described in Appendix A (Table A.1 and Figure A.1), and the full systematic search is available from the authors.    A priori, we operationally defined ‘workload’ as either external – the amount of work completed by the athlete (e.g. distance run, hours completed, etc.), or internal – the athlete’s response to a given external workload (e.g. session rating of perceived exertion, heart-rate based measures, etc.). We acknowledge that athlete self-reported measures often evaluate how athletes are handling training demands and may be referred to as ‘internal’ load measures, but we considered these perceptual wellbeing measures as a distinct step from quantifying athletes’ internal or external workloads [100]. Athletic injuries have been diversely defined in the literature, so we operationally defined athletic injury as any article that reported measuring ‘injury’, regardless of their specific definition (e.g. time loss, medical attention, etc.).  65   Two authors (JW + TG) screened the titles/abstracts of the systematic reviews. Where necessary, the full texts were retrieved to determine whether they should be included. A total of 6 systematic reviews [74–76,188,192,193] and 1 consensus statement [65] were identified that included at least 1 article meeting the inclusion criteria.   We extracted and reviewed the full texts of all the original studies included (n=279) in these 7 papers. For our analysis, we included all the original articles that met the following criteria:   1) Original articles were prospective cohort studies that examined the relationship between at least 1 measure of internal or external workload (as defined above) and athletic injury. Since theoretical models describe the recursive nature of injury risk with each training or competition exposure, workloads had to be continually monitored and include both training and match workloads for the same athletes. Although some athletes may have entered or left the group during the study period (e.g. through retirement or trades to other teams) the same team/group of athletes had to be followed throughout the study period, as opposed to repeated cross-sectional snapshots of different cohorts. 2) Articles collected intensive longitudinal data. We defined intensive longitudinal data as >20 observations per athlete [126].  3) Articles studied team sport athletes. We chose team sports because (1) there are high amounts of ILD collected in applied team sport settings [98], and (2) the majority of workload—injury studies are in team sport athletes (Jones, 2016). Military populations and 66  individual sports (e.g. distance running) were excluded due to the differences in task requirements and operating environment.  5.3.2 Patient and public involvement  As a methodological review, there was no patient or public involvement in this current investigation.    5.3.3 Article coding and description To describe the methodological, statistical, and reporting approaches utilised in each article, two authors (JW + CA) reviewed all the included papers and extracted 50 items of information for each article. These items included publication year, journal, variable operationalisation (e.g. internal vs. external load measures, injury definition, etc.), methodological approaches, statistical analyses implemented, reported findings, and more. To ensure consistency between coders, 10 articles were randomly selected, coded independently by both reviewers, and compared to assess agreement. Discrepancies were discussed by the two coders and an additional 5 articles were randomly selected and coded independently. The remaining articles were coded by JW and checked by CA.   5.3.4 Assessing how statistical models aligned with Collins’ threefold framework  To evaluate the statistical approaches used in this field, we first identified the key themes and challenges within the theoretical models and temporal design features within the workload—injury field, then developed a qualitative assessment to evaluate the statistical approaches.   67 Collins’ component 1 – The theoretical models that underpin athletic workloads and injury risk (in brief)  Briefly, we identified at least 3 key elements of athletic injury aetiology models. First, sports injuries are multifactorial [36,69,71]. Aetiology models since 1994 have all explained between-athlete differences in injury risk by identifying a host of ‘internal’ (e.g. athlete characteristics, psychological wellbeing, previous injury) and ‘external’ (e.g. opponent behaviour, playing surface) risk factors. More recently, Meeuwisse et al.’s dynamic recursive model [4] and the workload—injury aetiology model [131], have highlighted the recurrent nature of injury risk, meaning each athlete’s injury risk (i.e. within-athlete risk) also fluctuates continually as they train or compete in their sport (Figure 3.6). Thus, a second theme is that injury risk differs between- and within-athletes.  Finally, more recent injury aetiology models have highlighted injury risk as a complex, dynamic system (Figure 5.1) [180,194]. Complex systems, as in weather forecasting or biological systems [195,196], possess many key features, including an open-system, inherent non-linearity between variables and outcomes, recursive loops where the system output becomes the new system input, self-organisation where regular patterns (risk profiles) may emerge for given outcomes (emergent pattern), and uncertainty [180].68    Figure 5.1: Complex systems model of athletic injury.   Web of determinants are shown for an ACL injury in basketball players (A), and in a ballet dancer (B)  69 Collins’ component 2 – Temporal design / data collection The theoretical models relating workloads and injury illustrate a continuously fluctuating injury risk, with many variables that influence risk on a daily or weekly basis [4,131,180]. Thus, if researchers want to investigate the association between workloads and injuries, these data must be collected frequently enough to observe changes in these variables as they occur (temporal design). With technological advances, athletes’ physiological, psychological and physical variables are now often collected on a daily, weekly or monthly basis, along with ongoing injury surveillance data [96,187]. Therefore, in the workload—injury field, the theoretical models (injury aetiology models that describe regular fluctuation in workloads and injury risk) and the temporal design (frequent, often daily, data collection) are often well-aligned, especially in prospective cohort studies using ILD. This leaves us to consider only whether Professor Collins’ third component – the statistical model – aligns with these first two. Collins’ component 3 – Statistical model   From the theoretical aetiology models underpinning the workload—injury association, we highlighted three key themes to consider when choosing a statistical model: (1) injury risk is multifactorial, (2) between-athlete and within-athlete differences in injury risk fluctuate regularly, and (3) injury risk may be considered a complex, dynamic system.   From a temporal design perspective, intensive longitudinal data (ILD) are necessary to address these key themes, but they also carry at least 5 challenges that influence the choice of the statistical model.  70  1) Differentiating between- and within-person effects.  2) ILD include time-varying variables (e.g. workloads) and may also incorporate stable (time-invariant) variables (e.g., sex).  3) The ‘dependency’ created by repeated measurements of the same individuals violates the assumption of ‘independence’ common to many traditional analyses [127,197]. 4) Almost all longitudinal datasets have missing or unbalanced data [126].   5) Longitudinal data analysis require researchers to consider the role of time in their analysis [186].   5.3.5 Evaluating statistical approaches  We deliberately tried to align components 1 and 2 of Collins’ framework by describing the theoretical models underpinning the workload—injury association and only including articles that had a temporal design characterised by ILD. To review whether statistical approaches aligned with these two components, two authors (JW + BZ) qualitatively assessed whether the statistical models, as employed in the included studies, (1) were multifactorial, (2) differentiated between- and within-athlete differences in injury risk, and (3) analysed the data as a dynamic system – the three themes highlighted in the theoretical framework. From the temporal design, the same two authors evaluated whether the statistical analyses (4) included both time-varying and time-invariant variables, (5) were robust to missing/unbalanced data, 6) addressed the dependencies created by repeated measures, and (7) incorporated time into the analysis.  71  5.3.6 Data synthesis approach We first describe the characteristics of the included articles, then present our qualitative assessment of how well the various statistical approaches fit within Collins’ framework.    5.4 Results Thirty-four articles were included in this methodological review (Appendix A). In the first 10 articles coded by both reviewers, there were 10 discrepancies out of 500 total coded entries (10 papers x 50 items/paper), which gave us 98% agreement between reviewers. No item had more than 2 discrepancies. Of the 250 study criteria in the second set of 5 articles coded by both reviewers, there were 8 discrepancies (97% agreement).   Included articles were published from 2003 – 2016, with 78% of the studies published since 2010. Sports studied included rugby league (n = 10), soccer (n = 7), Australian football (n = 6), cricket (n = 5), rugby union (n = 2), multiple sports (n = 1), and basketball, handball, and volleyball (n = 1 each). Studies included an average of 96 athletes (median = 46), ranging from 12 [141] to 502 athletes [198]. The observation period for these cohort studies ranged from 14 weeks [157] to 6 years [136]. Most studies investigated male athletes (n = 30), with 2 studies on female athletes, and 2 on both sexes. Table 5.1 summarises the included articles’ basic characteristics, while the full data extraction table is available from the authors upon request.  . 72  Table 5.1: Summary of included workload—injury investigations, sorted by sport then publication year Reference Journal Study Length Sport n Level Sex (Rogalski et al., 2013) [140] J Sci Med Sport 1 season AFL 46 Elite Male (Colby et al., 2014) [138] JSCR 1 Season AFL 46 Professional Male (Duhig et al., 2016) [199] BJSM 2 Seasons AFL 51 Professional Male (Murray et al., 2017) [200] Scand J Med Sci Sports 2 Seasons AFL 59 Professional Male (Murray et al., 2016200 [201] IJSPP 1 season AFL 46 Professional Male (Veugelers et al., 2015) [139] J Sci Med Sport 15 weeks AFL 45 Elite Male (Anderson et al.,  2003) [141] JSCR 21 weeks Basketball 12 Sub-elite competitive Female (Dennis et al., 2003) [145] J Sci Med Sport 2 Seasons Cricket 90 Professional Male (Dennis et al., 2004) [202] J Sci Med Sport 1 Season Cricket 12 Professional Male (Dennis et al., 2005) [203] BJSM 2002-2003 cricket season Cricket 44 Sub-elite competitive Male (Saw et al., 2011) [204] BJSM 1 season Cricket 28 Elite Male (Hulin et al., 2014) [136] BJSM 43 indiv. seasons / 6 years Cricket 28 Professional Male (Bresciani et al., 2010) [205] Eur J Sport Sci 1 season (40 weeks) Handball 14 Elite Male (Gabbett, 2004a) [158] BJSM 3 years  Rugby league 220 Sub-elite Male (Gabbett, 2004b) [142] J Sports Sci 1 season Rugby league 79 Semi-professional Male  73   Table 5.1 continued: Summary of included workload—injury investigations, sorted by sport then publication year  Reference Journal Study Length Sport n Level Sex (Gabbett & Domrow, 2007) [159] J Sports Sci 2 seasons Rugby league 183 "Sub-elite" Male (Gabbett, 2010) [115] JSCR 4 years Rugby league 91 Professional Male (Gabbett & Jenkins, 2011) [143] J Sci Med Sport 4 years Rugby league 79 Professional Male (Gabbett & Ullah, 2012) [160] JSCR 1 Season Rugby league 34 Professional Male (Hulin et al., 2015) [113]  BJSM 2 Seasons Rugby league 28 Professional Male (Hulin, et al., 2016) [168] BJSM 2 Seasons Rugby league 53 Professional Male (Windt et al., 2016) [173] BJSM 1 season Rugby league 30 Elite Male (Killen et al., 2010) [157] JSCR 14 weeks Rugby league  36 Professional Male (Brooks et al., 2008) [198] J Sports Sci 2 seasons  Rugby Union 502 Professional Male (Cross et al., 2015) [52] IJSPP 1 Season Rugby Union 173 Professional Male (Arnason et al., 2004) [206]  AJSM 1 season Soccer 306 Professional Male (Brink et al., 2010) [207] BJSM 2 Seasons Soccer 53 Elite Youth Players Male (Mallo & Dellal, 2012) [208] J Sports Med Phys Fitness 2 seasons (2007/08 and 2008/09) Soccer 35 Professional (Spanish Division II) Male (Clausen et al., 2014) [209] AJSM 1 Season Soccer 498 Recreational  Female 74  Table 5.1 continued: Summary of included workload—injury investigations, sorted by sport then publication year  Reference Journal Study Length Sport n Level Sex (Owen et al., 2015) [210] JSCR 2 consecutive seasons Soccer 23 Professional Male (Bowen et al., 2016) [211] BJSM 2 Seasons Soccer 32 Elite Youth Players Male (Ehrmann et al., 2016) [169] JSCR 1 Season Soccer 19 Professional Male (Malisoux et al., 2013) [212] J Sci Med Sport 41 weeks Varied  154 High-school Both (65% males) (Visnes & Bahr, 2013) [213] Scand J Med Sci Sports 4 years (231 Student-seasons) Volleyball 141 Elite high school  Both (72F, 69M) 75  5.4.1 Data collection Injury definitions  Injury definitions varied across articles, with exact wording outlined in the Online Supplementary Material. In Table 5.2, we have categorised the definitions into more discrete injury categories (and subcategories) in accordance with recognised consensus statements [9].Where studies used multiple injury definitions, we categorised them according to the definition used for the primary analysis  Table 5.2 Broad injury definitions used in workload—injury investigations Injury Definition N   Time-Loss  All time-loss  13 Match time-loss 2 Non-contact time-loss 7 Non-contact match time-loss 1  Medical Attention  Medical attention 7 Player-reported pain, soreness, or discomfort 1 Non-contact medical attention injuries 1 Clinical diagnosis of jumper’s knee 1   Other  Injury scale on the Recovery-Stress Questionnaire for Athletes (REST-Q) 1  76 Subsequent or recurrent injuries Of the 34 articles, 30 did not define or include subsequent or recurrent injuries. Of those that explicitly addressed subsequent injuries, two defined these injuries as those occurring at the same time and occurring by the same mechanism  [208,210]. Two articles explicitly stated that they only considered time until first injury, meaning no injuries were subsequent or recurrent [145,203]. Workload definitions Workload variables varied widely across articles and are summarised in Table 5.3. For a more detailed description of each article’s load measures, see the Online Supplementary Material. Many articles used workload metrics to derive additional variables from workload distribution over time (e.g. monotony, strain, acute: chronic workload ratios).  Table 5.3: Independent variables used in workload—injury investigations Workload Measure  N  Internal  - sRPE  15 - Heart rate zones 2   External  - Balls bowled or pitched 5 - GPS / accelerometry  10 - Hours 6 * If articles included more than one type of workload variable they are counted more than once. sRPE scores could be the original Foster scale or modified. GPS – Global positioning system. sRPE – Session-rating of perceived exertion (calculated as the product of session intensity on a 1-10 Borg Scale and activity duration in minutes).  77 Measurement frequency Most included articles (n=32) collected workload data at every session that athletes completed, while 2 studies recorded workload on a weekly basis [209,213]. Handling missing data   Twenty-three of the 34 articles (67%) did not report any strategies for missing data. Of those that did, 5 used listwise or casewise deletion, and 6 used estimation. Estimation methods for players missing data included techniques such as: using the full team average values for the drills a player completed [211], using an individual’s mean weekly value [207], and multiplying player’s pre-season per-minute match data by the number of minutes they played in a match [138].   5.4.2 Statistical analysis and reporting in included articles Data binning/aggregation  Although 32 articles collected daily workload measurements, many aggregated data for analysis. Most (n = 16) summed workload metrics for a total or average weekly workload. Three studies aggregated workload data for the entire year, 3 aggregated data into season periods, 2 aggregated data monthly, and 3 used multiple aggregation strategies. Analysis methods Table 5.4 summarises the statistical practices of applied researchers investigating the relationship between workload and injury. Although some studies had analysed other primary or secondary objectives, we recorded only the analyses used to investigate the workload – injury relationship.  78  Table 5.4: The number of studies using various statistical analysis techniques Analytical Method N  Regression modelling   - Logistic   o Regular o Generalised Estimating Equation o Multilevel   10 5 1 - Linear  o Regular  2 - Poisson  o Generalised Estimating Equation**  1 - Multinomial regression o Regular  1 - Cox proportional hazards model 1 - Frailty model  1 Correlation - Pearson - Spearman  9 1 Relative risk/rate ratio* 8 T-tests - Paired and independent samples - Independent samples only  4 2 Chi-square tests 1 Repeated measures ANOVA (one or two-way) 5 If articles used more than one statistical method to analyse workload and injury, they are included more than once in the table. We only report analyses used to analyse workload—injury associations, not other analyses reported in the articles (e.g. ANOVA to test for differences in total workloads at separate times of the season). * Relative risk here refers to the use of RR as a primary analysis based on risks in different categorical groups, not as an effect estimated from another model. For example, comparing risks among different load groups like Hulin et al., (2014, 2016, 2016) are counted here, whereas Gabbett and Ullah (2012) derived RR from their frailty model, and Clausen et al. (2014) derived RR from their Poisson model, but neither are included in the count for RR.  ** Clausen et al. (2014) also report fitting multilevel models, but do not report any of the results – presenting only their GEE findings in their results and discussion.   79 Typical uses of statistical tools Regression approaches were used most commonly (22/34 studies). The most common approach was logistic regression (binary injury status as the outcome variable), independently or jointly modelling workload variables as independent variables. Generalised estimating equations were used in 5 studies to account for the clustering of observations within players and were used very similarly to simple logistic regression approaches.  Correlation was the second most common method (10/34 studies). Most studies that used correlation (7/10) measured the association between weekly or monthly workloads and injury incidence at the team level. Of those that used correlation at the individual level, two compared the number of completed pre-season sessions with the number of completed in-season sessions [131,200], while the final compared workload with injury operationalised as a numerical score on the injury subscale of the Recovery-Stress Questionnaire for Athletes (REST-Q) [205].   Relative risk approaches were generally used in one of two ways. First, workload categories were established for the entire year, like cricket bowlers who averaged <2, 2-2.99, 3-3.99, 4-4.99 or >5 days between bowling sessions up until an injury, or for the entire year if they did not sustain an injury [145]. Risks were calculated as the number of injuries/number of athletes in a given group, and relative risks were calculated to compare across groups [204]. In the second approach, athletes contributed exposures on a weekly basis, and thus contributed to multiple workload classifications. In this case, the likelihood/risk was the number of injuries/number of weekly player exposures to that workload category [136,168,211]. 80   Group differences were sometimes evaluated using t-tests, ANOVAs or chi-square analyses. Typically, unpaired t-tests contrasted workload variables (e.g. mean sessions/week) between athletes who sustained an injury during the year, to those who did not [145,203]. Paired t-tests and repeated measures ANOVAs (one- or two-way) were most often used to contrast the workloads of the same athletes at different time periods. For example, workloads in an ‘injury block’ (like the week preceding an injury), were contrasted with non-injury blocks, like other weeks in the season [145,204], or the 4 weeks preceding the injury block [212]. Justifications for statistical approaches Authors of 15 of the included articles (44%) did not cite any sources to support their analytical choices. Of those who did, most (n=14) cited previous literature in the sports medicine field. Eight articles referenced statistics or methodology articles, 4 cited articles on Prof. Will Hopkins’ website (, and 3 cited statistical textbooks [214–216]. Addressing analysis assumptions and model fit More than half (n = 20) the included articles did not report on the assumptions underlying their statistical analyses. Among those that did report on analysis assumptions, checks included checks for normality, collinearity of predictor variables in regression analyses [52], sphericity for repeated measures ANOVA [205], overdispersion [209], or correlation structures for generalised estimating equations [139].  81  When authors reported checking for normality, Shapiro-Wilk [200] or Kolmogorov-Smirnov tests [157] were referenced. Regression modelling was the most common analysis to investigate the workload—injury association. In 8/10 instances where simple logistic regression was chosen, the authors appear to have conducted the analyses using weekly observations without accounting for the dependencies created by repeated-measures across players. In all instances were regression was used, it was uncommon for authors to report that model assumptions were checked. Where multiple regression approaches were used, multicollinearity checks were rarely reported – an important consideration since multicollinearity can cause imprecise estimates of regression coefficients  when multiple workload variables are simultaneously modelled [52,217,218].  Of the papers that modelled data using regression or similar techniques, six described how they assessed model fit. Some authors assessed specificity/sensitivity, or receiver operating characteristics, either on the current data set [213], or future data set [115]. Other in-sample model fit indices R2 values [210], Aikeke Information Criteria (AIC) and Bayesian Information Criteria (BIC), which were sometimes mentioned as guiding the model selection process [173].  5.4.3 Alignment of authors’ statistical models with theoretical model and temporal design challenges In Table 5.5 (a more detailed table – Table A.2 - is available in Appendix A), we qualitatively evaluated whether the statistical approaches chosen by the authors in our current review effectively addressed the key themes/challenges presented by the theoretical model and the temporal design (intensive longitudinal data).  This table is an analytical tool to guide the reader through the discussion. It highlights the themes/challenges of the theoretical model and temporal design, as 82  well as the strengths/weaknesses of the statistical tools used in included studies. The table has the challenges/themes in columns and statistical tools in rows. The reader can follow a row to see how well a given statistical tool addressed key challenges as used by researchers in our included articles, or they can choose a challenge and follow the column down to see which analyses were used in a way that addressed that challenge adequately. The rows are ordered according to their qualitative ‘score’. As one proceeds down the rows, the statistical tools address more of the temporal design and theoretical model challenges.   We caution the reader that (1) not every possible statistical tool is included in the table, only those used in at least 1 article in our review, and (2) the evaluation is based on whether researchers of our included papers used a test in a way that addressed a given challenge, not necessarily whether the test is capable of being used in a way that meets that challenge. For example, a logistic regression analysis conducted using a generalised estimating equation framework can include multiple explanatory/predictor variables, thereby allowing for a multifactorial model. However, some authors used GEEs and only included one predictor variable [159,202], in which case the GEE did not address the multifactorial theme.   5.5 Discussion We used the workload—injury field of medical research to examine whether statistical approaches analyse intensive longitudinal data optimally. By design, the theoretical models underpinning the workload—injury field and the temporal design (ILD) were aligned in all the included articles, but common statistical approaches varied in how adequately they addressed the key themes needed to align them with the other two components.  83   5.5.1 Consideration #1 – Theoretical theme – multifactorial aetiology Sports injury aetiology models of the last 2 decades have highlighted the multifactorial nature of athletic injury [69,71]. We asked whether the burgeoning body of research relating workloads and injury, is using modern statistical methods to capture workloads while incorporating known risk factors. Few articles in this review incorporated previously identified risk factors and workload into the same analysis. In some instances, the analytical approach prevented this from being an option. For example, simple analyses like t-tests, correlations, and chi-square tests do not allow for multiple variables to be included. In other instances, the statistical approaches allowed a multifactorial approach (e.g. generalised estimating equations) but researchers opted to focus on the effects of workloads in isolation [159,202].   Including known risk factors in workload—injury investigations is important from an aetiological perspective in at least two ways. First, failing to control for known risk factors may mean that key confounding variables are not included in the analysis and the relationship between workloads and injury are spurious. For example, women have a 2-6 times higher risk of ACL injury in soccer than their male counterparts [3,219]. If a study included both male and female soccer players and did not account for sex in the analysis, then differences in workload may be spuriously correlated with injury rates if male and female players performed varying levels of workload. Depending on the injury type and sporting group, previous injury, age, sex, physiological and/or biomechanical variables may all be important to include.   84  Secondly, by including additional risk factors into the analysis, the investigator may be able to identify moderation or effect-measure modification to better understand how risk factors and workload jointly contribute to injury risk [179,220]. As a reminder, there are subtle, but important differences between mediation, moderation, and effect measure modification that will influence analytical choices [221,222]. Effect modification occurs when the effect of a treatment or condition (e.g. a given workload demand), differs among different athlete groups. Interaction (or moderation), although similar, examines the joint effect of two or more variables on an outcome. Finally, mediation is concerned with the pathway of exposure to a given outcome, and what are potentially intermediate variables. Previously identified risk factors may aetiologically relate to workload in each of these three ways and may be explored through different modelling strategies. 85  Table 5.5: Evaluation of the degree to which authors’ use of statistical tools addressed theoretical and temporal design challenges     Themes of theoretical model  Themes of temporal design - intensive longitudinal data Method n Multifactorial aetiology Between and Within-Athlete Differences Complex System Includes Time-Varying and Time-Invariant Variables Missing/ Unbalanced Data*  Repeated Measure Dependency Incorporates Time into the Analysis Correlation (Pearson and Spearman) 10 X X X X X X X Unpaired t-test 6 X X X X X X X Chi-square tests 1 X X X X X X X Relative risk calculations 8 O X X X X X X Regression (logistic, linear, multinomial) 13 O X X X X X X Paired t-test 2 X X X X X   Repeated measures ANOVA  (one or two-way) 5 O O X O X   Generalised Estimating Equations  (Poisson and logistic) 6 O X X O   O Cox proportional hazards model 1  X X X    Multilevel modeling 1   X    X Frailty model  1   X     Qualitative assessment performed on a three-tiered scale. An ‘X’ (red formatting) means that none of the authors using this tool adequately addressed that specific challenge. In some cases, this may be because the statistical model was unable to address it, and other times it may be because of the way they used it.  An ‘O’ (yellow formatting) indicates that some authors addressed that challenge while others did not. This generally happened when the statistical tool could address a challenge but the authors sometimes chose not to use it in that way. A ‘’ (green formatting) indicates that all authors using this statistical tool addressed that challenge adequately. *Missing/unbalanced data here is that caused by intensive longitudinal data – meaning a different number of observations for each athlete during the observation period, some of which may be missing.86  Statistical approaches that allow multivariable analyses enabled researchers to examine the effects of workloads while controlling for known risk factors. Malisoux et al. (2013) used a Cox-proportional hazards model to control for age and sex while examining the effects of average training volume and intensity. Gabbett and Ullah’s frailty model (2012) incorporated previous injury – a proven injury risk factor – into the evaluation of the influence of different GPS workloads on injury risk. When investigating multifactorial phenomena, statistical approaches that enable multiple explanatory variables provide a more appropriate option.   5.5.2 Consideration #2 – Theoretical theme - between and within-athlete differences One of the primary benefits of ILD is that it enables researchers (when using certain analyses) to differentiate within-person and between-person effects [186]. In the sports medicine field, this would correspond to researchers asking, (1) why do some athletes suffer few injuries (between-person inquiry) while others appear ‘injury-prone’? and on the other hand, (2) at what point is a given athlete (within-person inquiry) more likely to sustain an injury? The simpler statistical approaches used by researchers in our included studies (correlation, t-tests, ANOVAs, regular regression) are limited in the number of variables they can include, and consequently cannot differentiate risk between- and within-athletes. Tests of group differences (independent sample t-tests and one-way ANOVAs) only differentiate between athletes (e.g. injured vs. uninjured), while repeated measures tests (repeated measures ANOVA and paired t-tests) only examine within-athlete differences (e.g. loads preceding injuries vs. loads during non-injury weeks).   Generalised estimating equations (GEE) were commonly used to address some of the longitudinal data challenges. Although this approach accounts for the clustering within-persons, it assumes the 87  effects of predictor variables are constant across all athletes [223]. Simple Cox proportional hazards models (Malisoux et al. 2013) are common in survival analyses, but do not differentiate between- and within-person effects [224].  Only two statistical tools were used in a way that examined between- and within-athlete differences in injury risk. The frailty model by Gabbett and Ullah (2012) modelled  each athlete as a random effect with a given frailty. The multilevel model by Windt et al., (2016) incorporated athlete-level variables (age, position, pre-season sessions) and observation-level variables (weekly workload measures). In the latter case, athletes’ weekly distances did not affect their risk of injury in the subsequent week (OR = 0.82 for 1 standard deviation increase, 95% CI = 0.55 to 1.21) – a within-athlete inquiry. However, controlling for weekly distance and the proportion of distance at high speeds, athletes who had completed a greater number of pre-season training sessions had significantly reduced odds of injury (OR = 0.83 for each 10 pre-season sessions, 95% CI = 0.70 to 0.99) – a between-athlete inquiry. These two examples highlight that certain analyses carry a distinct advantage of allowing researchers to tease out differences both between- and within- study participants   5.5.3 Consideration #3 – Theoretical theme - injury risk as a complex dynamic system Complex systems are defined, among other things, by the interaction between multiple internal and external variables that interact to produce an outcome. Simple analyses (t-tests, correlations), which cannot incorporate multiple variables, cannot examine the interaction between multiple factors. However, even other traditional analyses which are more effective in handling the 88  challenges of longitudinal data (e.g. generalised estimating equations, Cox-proportional hazards models) were not used to incorporate non-linear interactions between predictor variables.   The most recent reviews of athletic injury aetiology have highlighted complex systems models [180,225]. None of the analyses included in our review analysed intensive longitudinal workload—injury data with statistical analyses that fit within a complex systems framework. This lack of research may reflect the fact that the suggestion that injury aetiology fits within a dynamic, complex systems framework is still relatively ‘new’ in this field. It remains to be seen whether a complex systems approach and the analyses recommended in such reviews (e.g. self-organising feature maps, classification and regression trees, agent based models, etc.) are more effective for evaluating the association between workloads and injury [180].    5.5.4 Consideration #4 – ILD challenge – including time-varying and time-invariant variables Tying back to the theoretical model of workloads and injury, some relevant factors may be relatively stable (time-invariant) over the course of an observation period (e.g. height, age), while others are time-varying (e.g. workload). Some analyses can incorporate both time-varying and time-invariant variables, while others are limited in this respect. All analyses that cannot or did not address the multifactorial nature of injury cannot include time-varying and time-invariant variables concurrently. Group difference tests (t-tests, ANOVAs, etc.) may collect time-varying measures, but must aggregate them into a single average for analysis.   89  Including time-varying and stable variables in the same analysis links closely to between- and within-athlete differences – with the frailty model [160] and multilevel model [173] both used in a way that allowed the researchers to include both. The one exception in our included studies was the generalised estimating equation approach. As mentioned earlier, the GEE assumes an ‘overall’ effect for each explanatory variable, such that between- and within-athlete differences cannot be differentiated [226]. However, the major benefit to a GEE is that it accounts for the repeated measures for each participant and can therefore include both time-invariant and time-varying variables for each participant.    5.5.5 Consideration #5 – ILD challenge – handling missing and unbalanced data Dealing with missing and unbalanced data is a near certainty when collecting ILD, and is common in applied workload-monitoring settings [227]. Such missing data decreases statistical power and increases bias, and may be missing at completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). When analysing aggregated data or using analyses that require balanced data, strategies may include complete-case analysis, last observation carried forward, or various imputation methods [228,229]. Multiple imputation methods, of which there are many, involves replacing missing values with values imputed from the observed data and is preferred over single imputation. Finally, if interactions are included in regression analyses, the transform-then-impute method has been recommended [230].   However, these missing data approaches are not recommended for longitudinal analyses, since researchers have statistical analyses that are robust to missing and unbalanced data at their disposal [128]. Statistically, four types of analyses used in this review are robust to missing and unbalanced 90  data – Cox proportional hazards models, GEEs, multilevel models, and frailty models, where all observations can be included in the analysis, and athletes can have different numbers of observations. Since mixed/multilevel models have less stringent assumptions for missing data (i.e. missing at random) than generalised estimating equations (i.e. missing completely at random), they have been suggested over GEEs [128].    While the statistical concerns related to unbalanced data may be addressed with these analyses, missing data may also affect derived variables, which are common in workload—injury research. These derived variables include rolling workload averages (e.g. one-week, ‘acute’, workloads, four-week average, ‘chronic’, workloads, etc.) [136,211], ‘monotony’ (average weekly workload divided by the standard deviation of that workload) or ‘strain’ (the monotony multiplied by the average weekly workload) [141]. Since these measures are all calculated from workloads accumulated over time, failing to estimate workloads for these missing sessions (that end up being treated as ‘0’ workload days) means inferences from these derived measures may be underestimated and unreliable. Few authors discussed how they handled missing data. In these instances – it is important that researchers report how they accounted for missing data, whether they be strategies employed in the past – e.g. full team average values [211], weekly individual averages [207], player specific per-minute values by time played [138] – or whether through other advanced imputation methods recommended for ILD [228,230].   5.5.6 Challenge #6 – ILD challenge – dependencies created by repeated measures Collecting ILD in applied sport settings means repeated (often daily) measurements of the same athlete – such that observations are clustered within athletes. Comparisons of independent groups, 91  through chi-square tests, independent sample t-tests and one-way/two-way ANOVAs all assume participants contribute a single observation to the analysis and force an aggregated variable (e.g. average number of balls bowled in a week) to conduct the analysis [145]. Similarly, correlation and simple regression (in its linear, logistic, and multinomial forms) assume independence of observations [231]. Paired t-tests and repeated measures ANOVA were used to deal with repeated measurements by comparing the same athletes’ workloads at different periods (e.g. the week before injury vs. weeks that did no precede injury).   Of the analyses that addressed this challenge, GEEs were used most commonly ( 6 studies). GEE’s ability to handle clustering was also used in one article to control for players clustering within teams [209]. Cox proportional hazard models, used in one article [212], can handle repeated measurements for participants [232]. Multilevel models [173] and frailty models [160] – an extension of the Cox- proportional hazards model – were also used in a single instance each, where repeated measures were clustered within players through a random player effect.   As mentioned in our introduction, there may be additional data dependency created by recurrent injuries [38]. Previous recommendations to handle the recurrent injury challenges have included frailty models [44], and a multistate framework [40]. However, as so few articles reported collecting information on recurrent injuries (n = 5), we focused primarily on the dependencies caused by repeated measures across participants.   92  5.5.7 Challenge #7 – ILD challenge – incorporating time into the analysis / temporality  One of the most relevant questions in ILD analyses is the way that time is accounted for [130,186]. Some authors used one-way [157] and two-way repeated measures ANOVAs [158] to compare loading in different seasons or season-periods – a very simple way of accounting for time. Repeated measures ANOVAs [169,200,212] and paired t-tests [145,204] also account for time by categorising time-periods as pre-injury blocks or non-injury blocks. Multilevel models have been used to examine change through the interactions of variables with time, but the one multilevel model used in this review did not include time as a covariate [173]. Survival analyses explicitly account for time by calculating the effects of variables on the predicted time-to-event [160,212]. Notably, only one analysis – the frailty model ([160]) – adjusted the probability of long-term outcomes (e.g., injury) based on variations after an initial capture of risk, something few traditional analyses accomplish [195].   Temporality is also vital in considering potential causal associations. While making causal inferences from observational data is a topic beyond the scope of this paper, temporality is a well-accepted component of causality dating back, at least, to Bradford Hill’s ‘criteria’ [233]. Without temporality – where a postulated cause precedes the outcome – directional associations cannot be made [234,235]. A lack of temporality can also skew associations since it allows for reverse causality. In the workload—injury field, findings that high weekly workloads are sometimes associated with lower odds of injury in a given week [139,173] may be in part because players who get injured in a given week are less likely to accumulate high weekly workloads.  Trying to account for temporality, some researchers have included a latent period – where workload variables are examined for their association with injury occurrence in a given proceeding 93  time window, like the subsequent week [136,173]. While recent work has noted that the length of the latent period may affect model findings [236], it is clear that without some type of latent period, any directional inferences between workloads and injury cannot be made.  5.5.8 Methodological, statistical and reporting considerations Data aggregation Data aggregation was common, whether in data preparation, or forced through the analysis. In some cases, researchers aggregated individual level data into team-level measures (total/average workload and injury incidence). Although 32/34 articles collected daily data, most aggregated these daily data into weekly measures, potentially contributing to temporality problems if no latent period was included. Finally, certain analyses (e.g. paired t-tests, simple logistic regression) aggregated data for athletes across an entire year so that workload measures were used to control for exposure [206,213]. Differences in analyses make it impossible to measure the effect of fluctuations in workload and potential impact on injury risk. Further, with no latent period, the directionality of the relationship is unclear. For example, players with high exposure throughout the year were at a lower injury risk than the intermediate group, but it could be interpreted that players who do not sustain an injury throughout the year are more likely to accumulate high total training and match exposures (i.e. higher workload) [206]. Aggregated data may be easier to analyse but comes at the cost of losing some of the inherent benefits of collecting ILD, such as the changes in injury risk that occur at a daily level. As a result, theory-driven questions that relate to daily workload fluctuations and injury risk will become challenging, or impossible to answer.   94 Checking model assumptions and fit While many studies may have under-reported how they assessed model assumptions or fit, others [52] provide an example for other researchers to emulate. In fitting a GEE to account for intra-team and intra-player clustering effects, they explained how they selected an appropriate autocorrelation structure, reported how potential quadratic relationships were assessed in the case of non-linear associations, and described checking for potential multicollinearity with defined thresholds (variance inflation factor >10) and for their GEE.   5.5.9 Researcher ‘trade-offs’, consequences of misalignment We used the workload—injury field to highlight seven themes that relate to theoretical injury aetiology models and temporal design (ILD). In many cases, published studies’ statistical models either could not, or were not used in a way that addresses these themes. In some cases, misalignment may carry a severe cost – like assumption violations that may bias study results [197]. This is akin to building conclusions on an unstable foundation. Other times, researchers have properly employed their chosen statistical approach, but the approaches themselves were limited, and unable to answer research questions that ILD can address. This is more akin to having a grand building plan and all the necessary supplies, but only using a screwdriver to construct the building.   Simple regression models provide an ideal example of researchers’ trade-offs when using traditional statistical analyses on ILD, and the potential costs of misalignment. Although 13 papers used regular regression to analyse the association between workloads and injury outcome, they chose one of three paths when dealing with ILD. First, many proceeded to analyse each daily or 95  weekly data point as an independent observation – not addressing the violation of the independence assumption [136,140,168]. Second, some researchers aggregated the workload data into an average weekly workload or total workload exposure over the course of the year, such that each participant contributed only one observation to a classic logistic regression [206,213]. Although the regression assumptions were not violated, workload was aggregated into a single metric, the temporal relationship between workload and injury was lost, and there was then no way to analyse the effects of workload fluctuations on injury risk. Third, some researchers converted individual data to team level data and examined team workloads with team injury incidence in a linear regression [142,210]. In this final case, no differentiation could then be made between players or within-players, and inferences were only possible at the team level. This may be sufficient to inform research on the association of workloads and injury at the team level, but the theoretical model underpinning team injury rates may differ from those that underpin individual athletes’ injury risk.  5.5.10 Review limitations Previous systematic reviews investigating the workload—injury relationship have documented the challenges of identifying articles through classic systematic review search strategies [74,76]. Heterogeneous keywords and the breadth of sporting contexts have meant previous systematic reviews include many articles post-hoc that were not originally identified by their systematic searches (e.g. 29 of 67 articles in Jones [76], 12 of 35 articles in the paper by Drew and Finch [74]). Therefore, although we worked to identify articles through 6 systematic reviews [74–76,188,192,193] and the 2016 IOC consensus statement on athletic workloads and injury [65], we may have missed potentially eligible articles .   96  We used the cut-off for intensive longitudinal data (>20 observations) proposed by Collins (2006). However, there is no universal cut-off for ILD, with previous thresholds of ‘more than a handful’ [130], ten observations [237], or forty [186].   In some instances, authors’ analytical choices may have been attributable to factors outside of statistical considerations. For example, in lower level competitions, or in organisations with lower budgets, it may not have been feasible to collect multiple variables longitudinally with the available equipment or staff. In these types of instances, authors would be unable to employ a multifactorial approach, instead of choosing not to use one. Such external factors may have influenced the findings of this methodological review.   Finally, it was beyond the scope of this review to list every challenge posed by ILD, and we were not exhaustive in our discussion of different analyses and their capacity to handle the challenges. Where possible, we tried to identify the themes that are most common within the research field of Sport and Exercise Medicine field. Ultimately, our call to action is that statistical tools be chosen more thoughtfully so that the extensive work put into theory building and data collection is not short-changed by a sub-optimal statistical model.   5.5.11 Longitudinal improvements in ILD analysis Methods and statistical analyses evolve over time, as with all scientific inquiry. Therefore, it is possible that we were a little unfair to some earlier papers. For example, researchers may have chosen analyses that aligned with ‘their’ theoretical model at the time, not what is considered the most current theoretical model. However, most papers were published since 2010 – the dynamic, 97  recursive aetiology model was introduced in 2007, and the multifactorial nature of injury risk has been highlighted since 1994 [69]. As complex systems approaches are the most recently proposed theoretical model [180,225], it is not surprising that none of the included articles analysed the data within this type of framework, with the first analysis of its kind in sport injury research only appearing recently [238]. Further, some techniques for longitudinal data analysis have been developed and grown in popularity recently, so researchers may not have been aware of alternative approaches at the time of their studies.   As more statistical methods are developed and refined for longitudinal data analysis, researchers will continue to gain awareness and skills with these analyses and their implementation is likely to become more common. Some evidence for that progression can be seen in this review. If we were to assign a ‘method’ score to each analytical approach outlined in Table 5.1, assigning 0 for each red box, 0.5 for each yellow box, and 1 for each green box (e.g. correlation would score 0, while generalised estimating equations would score a 3.5), and then assign that score to each paper in the study, we could obtain a rough estimate of whether analytical approaches were improving over time. Breaking the papers roughly into four periods, the ‘average score’ for papers up to 2005 (n = 6) is 1.6, papers between 2006 and 2010 (n = 7) score an average of 1.9, papers between 2011-2015 (n=11) score 1.7, and papers since 2016 (n = 10) score an average of 2.3. Moreover, since the search for this current review was conducted, there have been promising developments in the sports medicine field and a continued improvement in longitudinal analysis. Recent publications have applied statistical models that more appropriately take advantage of the strengths inherent to ILD, and better align with the theoretical frameworks [239–244].   98  Mediation, effect measure modification, and interaction/moderation are all causal models which may also contribute to aetiological frameworks [181]. We recently proposed that traditional intrinsic and extrinsic risk factors may act as moderators or effect measure modifiers of the workload—injury association [179]. If that is true, the most appropriate statistical model would include workload measures as the independent variable of interest, and incorporate other risk factors such that these causal models can be investigated, whether by stratifying effects across different levels of these risk factors, or including an interaction term within regression [220]. While no included articles performed such an analysis, recent studies (not included in this review because it was published after our search) have started to adopt these approaches [241,245,246]. For example, Møller et al. used a frailty model with weekly workload fluctuations (decrease or <20% increase, 20-60% increase, and >60% increase) as the primary predictor variable in a frailty model. Known shoulder risk factors were treated as ‘effect measure modifiers’, so the model was stratified based on the presence or absence of a given risk factor (e.g. scapular dyskinesis) [222]. In so doing, the researchers used a statistical tool (Component #3) that addressed all the challenges inherent to longitudinal data (Component #2), conducting a multifactorial analysis that clearly differentiated both within- and between-athlete injury risk – key aspects of the theoretical model (Component #1).   5.5.12 Future directions and recommendations for ILD analysis Researchers in the sports medicine field should be encouraged that the increased availability of ILD may improve understanding of athletes’ fluctuating injury risks – as articulated by their theoretical models. More advanced statistical techniques for longitudinal data are increasingly being developed and implemented across disciplines. This will enable sports medicine researchers 99  to more accurately answer their theory-driven questions by taking advantage of the benefits of ILD. To capitalise on this understanding, researchers must choose statistical models that most closely align with their theory and that address longitudinal data challenges. Generalised estimating equations, a Cox proportional hazards model, a multilevel logistic model, and a frailty model were the 4 analyses that most closely approached this alignment within our included papers. However, there remains some clear room for improvement in the future.    First, although mixed modelling was only used in one study, these forms of analyses have inherent values over GEE methods and have been recommended for this reason [247]. Because of sample structure, mixed models prevent false positive associations and have an applied correction method that increases the power of the analysis [248]; a finding that is useful with the commonly smaller samples. Mixed models also carry a less stringent missing data assumption (missing at random) when compared with GEEs (missing completely at random). Further, whereas GEEs require the correlation structure to be chosen by the researcher (which may be wrong), mixed models model the correlation structure so that it can be investigated. Finally, GEEs assume a constant effect across all individuals in the model, while mixed models allow for individual level effects and for differentiating these individual effects.   To borrow an example from another field and demonstrate the flexibility and utility of mixed effect models, Russell et al. used daily stressor values from students during their first 3 college years to demonstrate that students consumed more alcohol on high-stress days than low-stress days (within-person fixed effect) [249]. However, a significant random effect between students suggested that some students experienced this increase in alcohol consumption, while others did not. Finally, 100  those students with a tendency to increase alcohol consumption with stressors were more likely to have drinking-related problems in their 4th year [249]. For more information on multilevel/mixed effect models for longitudinal analysis, readers are referred to other helpful resources [127,128,130,250,251].  Time-to-event models are another family of statistical models that have become a very common in clinical research articles – reported in 61% of original articles in the New England Journal of Medicine in 2004-2005 [252] – but were used infrequently within our included articles. Notably, these models answer a different research question – when does an event occur? These approaches can account for many of the ILD challenges [253–255]. Time-to-event models account for censoring, can incorporate time-varying exposures, time-varying effect measure modifiers, and time-varying changes in injury status, and may be used to control for competing risks [253]. As with other modelling techniques, the appropriate number of events per variable has been investigated, and at least 5-10 events per variable are recommended for these types of models to prevent sparse data bias [256]. As long as this and other model assumptions are met, more advanced time-to-event models may be a valuable tool for researchers analysing ILD [232,257,258].  Lastly, computational modelling methods, which involves computer simulation has both pros and cons where modelling injuries. On one hand, they may provide insight on the best ways to model certain predictor variables [259], and open the door to more complex systems modelling (e.g. agent-based modelling) [238]. Though they show promise, such simulation studies are based on artificially generated data and must be interpreted carefully [260].  101   More analytical approaches are available for ILD, but a full discussion of each of these is beyond the scope of this paper. For the interested reader, functional data analysis [261], machine learning approaches [239,242], time series analysis [129], and time-varying effect models [262] all show promise. Such analyses and others for ILD can be found in Walls and Schafer’s landmark ILD textbook [130], and more recently, in the work of Bolger and Laurenceau [250].  We believe intensive longitudinal data provide an exciting opportunity for applied researchers and statisticians to collaborate moving forward. As the field continues to progress to more advanced analytical approaches that may better suit ILD, the need for collaboration with statisticians will be vital. In our included papers, few researchers referenced methodological or statistical references to justify their analytical approaches. In some instances, this may be attributable to using common, relatively simple analyses – one likely does not expect a citation for a t-test. Where such references existed, they were often to previous papers in the field, not statistical sources. In future longitudinal analyses, we encourage researchers to partner with a statistician, psychometrician, epidemiologist, biostatistician, etc. [263]. Such fruitful collaborations may lead to statistical approaches that take full advantage of intensive longitudinal data by aligning theory, data collection and statistical analyses as seamlessly as possible.   5.6 Conclusion We used studies investigating the relationship between workloads and injury as a substrate to highlight to researchers how important it is to align their theoretical model, temporal design, and statistical model. In longitudinal research, thoughtfully chosen statistical analyses are those 102  grounded in subject matter theory and that maximise the utility of the collected data. The three most common analyses in our included papers (logistic regression, correlations, and relative risk calculations) addressed one or none of the 3 key theoretical themes, and one or fewer of the 4 inherent challenges of intensive longitudinal data. In this example discipline, researchers have developed sophisticated theories and frequently collect data that enable them to test these theoretical models. The missing step, and future opportunity for researchers, is to avail themselves of all the tools at their disposal – choosing statistical models that address the ILD challenges and answer theory-driven research questions.  103  Chapter 6: Does player unavailability affect football teams’ match physical outputs? A two-season study of the UEFA champions league (Paper 4) 6.1 Preamble In Chapter 5, I reviewed 34 original prospective cohort studies that analysed intensive longitudinal data to investigate the workload—injury association. Authors selected generalised estimating equations most frequently to account for the dependency created by repeated measurements in these studies. Multilevel (i.e. mixed) modelling approaches have distinct advantages over GEEs, as described in section 5.5.12, and only one of the articles used a mixed modelling framework (Chapter 8).   In this chapter, I demonstrate the utility of mixed modeling using an intensive longitudinal data set. The investigation, which I carried out in partnership with the Football Research Group1, altered the explanatory and outcome variables – asking whether teams’ player unavailability (because of athletic injuries) was associated with match physical outputs (e.g. distance run, number of high velocity sprints). The repeated matches played by each team exposed the analytical challenges highlighted in Chapter 5, with match observations clustered within each team. Given the benefits of mixed modelling approaches over GEEs and other analyses ill-suited for ILD, I employed a mixed model to investigate whether player unavailability was associated with any changes in teams’ match physical outputs.                                                   1 Football Research Group (FRG) is a collaborative, international research group anchored at Linköping University, partnering with the European Football Union and elite football clubs, including Real Madrid, FC Barcelona, Juventus and Manchester United. Since its inception in 2001, 55 clubs from 18 countries have participated. The FRG database contains over 22,000 injuries and over 2 million hours of exposure, making it the world’s largest database on male professional football players.  The FRG’s goal is that exercise and football participation should be safe, with minimal injury risk and side effects.   104  The rest of this chapter is a reproduction of text and figures from the following article [264]:  Windt, Johann, Jan Ekstrand, Karim M. Khan, Alan McCall, and Bruno D. Zumbo. “Does Player Unavailability Affect Football Teams’ Match Physical Outputs? A Two-Season Study of the UEFA Champions League.” Journal of Science and Medicine in Sport, August 24, 2017.  6.2 Introduction Injuries are common in professional football [265,266] and negatively affect team performance [53,56,134]. While the link between injuries and team performance has been established [56], the mechanisms through which injuries affect teams’ match play and subsequent performance has not been investigated. Given the multifaceted nature of team sport performance, having more players unavailable may cause teams to alter their tactics, as less skilled players are selected may result in worse technical proficiency and altered physical performances during match play. While all these avenues provide future research opportunities, we will focus on the association between player unavailability and football teams’ match physical outputs.   Individual players’ physical outputs – including their total distance run, high speed distance run, number of sprints performed, time spent at different velocities, number of repeated sprint demands, and many other variables – are now readily quantified. This has enabled researchers to examine intra- and inter-individual variation over successive matches [267–269], as well as the effects of player position [270], team formation [271], team ranking [272], and fatigue (transient and end-game) [273]. To our knowledge, whether player availability due to injury is related to any changes in football teams’ physical outputs has not been examined.  105   If player unavailability is associated with changes in match physical outputs, sports medicine and science staff may benefit in two ways. First, the impact of injuries may be more readily quantified – improving preventative efforts. Second, teams with more players unavailable for selection may better understand what to expect in coming matches and therefore prepare appropriately. Our objective was to investigate whether player availability was associated with any changes in teams’ match physical outputs. Theoretically, having more players unavailable due to injury may cause less ‘match fit’, or more inexperienced players to be selected for match play, and reduce teams’ ability to utilise squad rotation during times of fixture congestion. Therefore, we hypothesised that player unavailability would be associated with a decrease in teams’ physical outputs.   6.3 Methods The UEFA Champions League (UCL) injury study is an ongoing prospective injury surveillance study of male professional football that started in 2001 [274]. The current study cohort followed the teams participating in the UCL injury study who competed in the 2014-2015 and 2015-2016 UEFA Champions League competitions. Only UCL matches were included, without any league matches. UCL fixtures are generally played every 2-3 weeks on Tuesdays and Wednesdays between September and May. Teams play 6-12 matches depending whether they advance to the knockout stage and how long they remain in the competition.  Data collection procedures and definitions followed the UEFA guidelines [274]. and were consistent with the consensus statement for injury surveillance in football [9]. All players 106  completed informed written consent and the study design was approved by the UEFA Medical Committee.   First teams’ medical staff members logged all injuries on a standard injury form, sent to the study group each month. A time-loss injury definition was used for this investigation, with injuries recorded whenever a player’s physical complaint prevented full football training or match play [9]. Players were considered injured until the medical staff allowed full participation in training and availability for match selection. The number of players unavailable due to injury was calculated for each match as the ‘count’ of how many players were designated as ‘injured’ on the date of the match and therefore unavailable for selection. Players unavailable due to international duty, yellow cards, or other reasons were not included in this count.  Performance data were collected through video tracking and analysed by Deltatre. These are the same data Deltatre analyses for UEFA and reports to teams during and after matches [275]. The analysis algorithm used to calculate physical outputs was consistent across seasons, but the video tracking hardware provider changed between seasons – Stats in the 2014/2015 UCL Season and ChyronHego in the 2015/2016 Season. Therefore, we chose to analyse each season separately.  Physical output data included total distance run (km); distance run (km) and time spent (minutes) at low, medium, and high-speeds; number of sprints performed (n). Speed thresholds were set at < 0 - 10 km/h for low intensity, <10 – 14km/h for medium intensity, >14km/h for high intensity, and 20km/h for sprinting [267,276]. We specifically selected three physical output variables for this analysis – 1) total distance run, 2) number of sprints, and 3) percentage of total distance performed 107  at high speeds (>14 km/h). These outcome variables were chosen to examine both match volume (total distance) and match intensity (sprint count and % of distance above high speeds). Two matches were excluded due to technological errors (no sprints were recorded by the video tracking software), and 9 matches were excluded which went to extra-time, as the physical outputs of these matches were outliers for all analyses.  Both match and team characteristics were collected and included in the multilevel models as covariates. Match characteristics included ball possession (% of match time), score differential (e.g. -3, -1, +2, etc.), venue (Home/Away), and competition stage (Group Stage/Knockout). Team characteristics included average team age and team quality (determined by UEFA Club Coefficients).   Descriptive analyses and visualisation were completed in the open-source statistical software, R (Version 3.2.2), using the dplyr and ggplot2 packages. Multilevel models were constructed using HLM 7.01 for Windows [277].  Magnitude-based inferences for correlations and t-distributed statistics were calculated using a custom spreadsheet [278]. In accordance with previous approaches, we defined the smallest worthwhile change in each output variable as 0.2 times the between-team standard deviation, and as 0.1 for correlation coefficients. Finally, we examined the association between player absences and each outcome variable by analysing the change associated with a 2 standard deviation change in the number of players absent [279].  Results are expressed as means ± standard deviations. To examine the association between player unavailability and physical output, two approaches were used. First, across the season, between-108  team differences were calculated using Pearson correlation coefficients, examining the relationship between teams’ average player unavailability and their average physical outputs. Second, within-team effects of player unavailability were examined using multilevel (mixed/random effect) linear regression – modelling the three physical outputs as outcome variables, with player availability as the explanatory variable of interest.   Multilevel models were performed with match observations (Level 1) clustered within teams (Level 2). This modelling approach was chosen for its ability to handle the data dependency created by repeated measures of the same teams, the unbalanced data formed as teams participated in different numbers of matches throughout the competition [127,280], and recent recommendations for mixed models to be used in these types of analyses [281]. Following previous recommendations, we included all level 1 covariates as random effects, including level 1 covariates in raw metrics if they held a meaningful 0 point (score differential, 0 = draw), while centring other continuous variables (ball possession) at the group mean (a.k.a. centring within cluster) – recommended when trying to examine the effect of another Level 1 variable [282]. The full model building strategy can be seen in Appendix B, and all the multilevel model outputs are included as supplementary material. Where sample size allowed, we used the robust standard errors calculated in HLM.   6.4 Results In 2014-2015, 23 teams were included with 180 match observations.  In 2015-2016, 20 teams were included, with 158 match observations. Fourteen teams participated in both UEFA seasons (total 109  n = 29). The breakdown of teams’ demographic information as well as injury breakdown is included in Appendix B.   Player unavailability ranged from 0 to 8 players unavailable due to injury on match day, with an average of 2.1 players unavailable in 2014/2015 (SD = 1.91) and 2.7 players in 2015/2016 (SD = 1.84). Rounding these standard deviations to the nearest integer (2), we calculated the changes in our outcome variables associated with 4 players unavailable for selection (2 x SD) [283]. The distribution of player unavailability for each team across each month of play is shown in Appendix B (Figure B.1).   The average physical outputs as well as the within and between-team variability of various physical outputs are recorded in Table 6.1.  Due to the differences in physical outputs between the two seasons, remaining analyses were conducted with each season independently.   110  Table 6.1: Average and variability of teams’ physical outputs in the 2014-2015, 2015-2016 UCL seasons  2014/2015 Season 2015/2016 Season Physical Output Measure Mean* Range* Within-Team SD** Between-Team  SD*** Mean Range Within-Team SD Between- Team  SD Total  Distance (m) 113 451  99813 – 125 086 3898.59 2566.35 107 843 94 234 – 119 748 3778.76 3014.28 High Speed  (14-20 km/h) Distance (m) 21 765 16 660 – 6 537 1487.41 1256.47 19924 14 967 – 25 377 1720.96 1214.23 Very High Speed (>20km/h) Distance (m) 10285 6917 – 15 108 1278.45 687.54 8760 6421 – 11 039 911.66 406.87 Sprints  (number of efforts >20km/h) 417 266 – 562 50.27 26.64 341 239 – 442 40.40 17.69 % of Total Distance Performed at High + Very High Speeds (>14km/h) 28.2 23.3 - 33.1   1.52 0.98 26.6 21.7 – 31.0 1.62 0.72 * Average and ranges calculated for all match observations.  ** Within-team standard deviation is calculated as the standard deviation of the level 1 residuals.  *** Between-team standard deviation is calculated as the standard deviation of the random intercept term in a null (random intercept only) model with no independent variables.  111  Teams’ average player unavailability and average physical outputs were correlated for each season. Figure 6.1 displays the results from each of these analyses. Likely positive relationships existed between average player unavailability and average number of sprints per game for both seasons (Pearson’s r = 0.38 and 0.37, respectively) and a likely positive increase was seen in % of distance at high speeds with player absences in the 2014/2015 Season. All other associations were unclear.  Estimated associations of player unavailability and each of the 3 physical outcomes, controlling for match and team characteristics, are presented in Figure 6.2. Having 4 additional unavailable players unavailable for selection was associated with an estimated increase of 20.8 sprints in 2015/2016 (p = 0.040), and an 0.60% (p = 0.035) and 0.77% (p = 0.028) increase in the percentage of distance ran at high speeds (>14 km/h) in 2014/2015 and 2015/2016, respectively. There was a ‘possibly positive’ and ‘likely positive’ association between player unavailability and total distances, with estimated increases of 597 meters and 1395 meters for 4 additional unavailable players (Figure 6.2). All qualitative inferences were positive, except for an unclear association between player unavailability and sprint count in the 2014/2015 Season. A summary table of model findings can be found in Appendix B, and full HLM model outputs are available as Online Supplementary Material. HLM model assumptions were examined through residual plots and no issues warranting considerations were found.  112   Figure 6.1: Teams’ average player unavailability for each season and their average: 1) Distance per game (m), 2) Number of sprints (efforts >20km/h) per game, and 3) Percent of total distance covered above 14km/h per game. 113   Figure 6.2: Estimated effects of 4 additional unavailable players on physical output measures  Row 1 = total distance (m), row 2 = sprint count (efforts above 20km/h), and row 3 = percentage of total distance above 14km/h. Shaded area represents the smallest worthwhile change in the physical output (calculated as 0.2 times the between-team standard deviation for the physical output variable). For interpretability, effect estimates were calculated as the 90% CI for the effect of 4 additional unavailable players. Magnitude-based inferences were calculated to interpret the probabilities for this effect as negative, trivial or positive from a  custom spreadsheet [284].   * Centre value is the model intercept – estimated physical output for a team playing a group stage match wherein the venue is away, teams have their average ball possession time, and the match finishes as a draw.   114  6.5 Discussion Player unavailability was associated with increased match intensities in the 2014/2015 and 2015/2016 seasons. First, teams’ average player unavailability was positively correlated with their average match sprint counts in both seasons, and with the percentage of distances covered at high speeds in the 2014/2015 season (Figure 6.1). However, since correlation does not equal causation, and these were averaged over the whole season, we cannot determine the direction of this relationship. Increased sprint counts may contribute to increased injury counts, or increased injury counts may contribute to increased sprint counts. It could also be the case that both effects could be present. Some evidence to suggest that player unavailability influences match intensities is presented in the multilevel models, since they account for player unavailability entering match play, controlling for match and team characteristics. These models indicated that higher player unavailability was associated with an increased percentage of distance covered at high speeds in both seasons, and with an increased sprint count in the 2015/2016 Season (Figure 6.2).   Player unavailability was also associated with ‘possibly’ and ‘likely’ positive increases in match distances (Figure 6.2). The fact that these associations were not statistically significant likely stems from the multifaceted nature of football match play – effected by team and opponent strategy, environmental conditions (heat/altitude), match importance, match outcome, and many more factors [281,285,286]. Controlling for as many of these factors as possible in our current modelling approach, player unavailability may also be associated with increased match volumes.   These results were contrary to our original hypothesis that there would be a negative association between player unavailability and physical outputs. Although currently speculative, 115  these findings may extend previous research comparing physical outputs between teams. Top-ranked teams engage in less high-intensity activities than lower than lower-ranked opponents [272]. Our results indicate that this may also be true within the same team, where increased player unavailability contributes to a relatively weaker line-up and concomitant higher match intensities. The reasons for these higher intensities remain unknown, but could be due to increased efforts to regain possession, or stem from decreased technical and tactical proficiency with ball possession, causing an increased reliance on physical outputs.  These findings should also be considered pragmatically within the context of professional football [271]. For example, if 4 players are unavailable for selection, teams would be estimated to run an additional 0.60-0.77% of their total match distances at high speeds, and as a team complete 21 additional sprints. Knowing what to expect during such matches when multiple players are unavailable may help sports medicine and coaching staff to more appropriately prepare for the upcoming fixture. For example, under-exposure to high-speed running increases athletes’ injury risk when they experience rapid workload increased, so ensuring a sufficient high speed running chronic workload may be important, especially for players with less frequent match exposures [114].  We analysed the association between player unavailability and match physical outputs at the team-level, across entire matches – not individual player outputs or outputs during different match segments. Examining individual player outputs and the time course of these outputs could provide more specific findings unaccounted for in our analysis. For example, although the intensity of teams’ physical outputs may increase with fewer players available for selection, we did not determine if those changes happened for certain players – like the players selected in place of injured athletes, or for certain positions (e.g. central defenders or fullbacks).  116   Similarly, although there was no relationship between match distance and player unavailability at the team level, it may be that the distribution of that distance varies across players or positions when fewer players are available for selection. Given the breadth of research that compares the physical outputs of players based on positions or physical fitness levels [270,287], future investigations should analyse the effects of player availability on player/position-specific physical outputs. Finally, investigating fatigue and pacing in elite team sports examine the time course of these match outputs – it may be that player unavailability is associated with an altered physical output distributions over time [273,281].   As highlighted previously, physical match outputs are multifaceted and we did not have all relevant variables collected in our current investigation (e.g. environmental conditions, team strategy, etc.). Knowing that player unavailability leads to poorer team performance [53,134], we focused on whether player unavailability was related to teams’ physical outputs. Whether a change in relative intensity or other physical output variable is involved in mediating the injury—team performance relationship is a question for further study. Future studies may also want to include measures of technical performance associated with match outcomes – like corners, passing percentage,  shots on goal, etc. [288] – or more advanced performance analytics [289], in conjunction with the variables we included. Incorporating such variables may assist in assessing the ways in which player availability effects match performances and team success.  6.6 Summary and conclusions As teams in the 2014/2015, 2015/2016 UEFA Champions League had fewer players available for match selection, they experienced increased match intensities – based on the estimated 117  percentage of total match distances run above 14 km/h and the number of sprints the team performed. Increased player unavailability was also associated with increased total match volumes. Knowing that match physical outputs may be increased when multiple players are unavailable for selection may equip sports medicine and coaching staff to better prepare their players for those fixtures.  118  Chapter 7: Is it all for naught? What does mathematical coupling mean for acute:chronic workload ratios?  (Paper 4) 7.1 Preamble Longitudinal workload monitoring has increased in prominence within the sports medicine and sport science field. The acute:chronic workload ratio (ACWR) is often at the center of these discussions – most often as an indicator of potential injury risk. To my knowledge, at least 26 original studies of workload—injury have incorporated the ACWR since it was first published under this name in 2014 (See Supplementary Material 1. Note Malisoux et al [212] used what would be an uncoupled ACWR, but didn’t refer to it as an ACWR). Although the ACWR has consistently been associated with injury risk in a range of sports, there are a range of opinions about its utility.  Editorials have criticized it [227,290–292], suggested modifying it [293,294], and supported its use [295–298].    Criticisms of the ACWR include: 1) the challenges of monitoring the ACWR in some professional team contexts – particularly football (soccer) [227], 2) using rolling averages as opposed to exponentially weighted averages [291,292], and 3) the ‘mathematical coupling’ that occurs (in the traditional 1:4 rolling average ACWR approach) because the most recent training week is included in both the acute and chronic loads [293].   These latter 2 criticisms are mathematical/statistical in nature, and fundamentally question the validity of using an ACWR in longitudinal monitoring. While Lolli and colleagues identified that mathematical coupling existed, with a clear correlation between the acute and chronic workloads in simulated data, they did not explicitly define what this correlation did to the ACWR calculation. Other than ‘uncouple’ the acute and chronic periods, they also did not 119  unpack how practitioners and researchers should respond to ‘mathematical coupling’. Therefore, my purpose in this paper was to answer these two questions. First – what does mathematical coupling between the acute and chronic workloads do to the ACWR calculation. Second, what does this mean for applied sport science/medicine research and practice. My aim was to bridge the mathematical discussion with the applied research and practice realm.    The rest of this chapter is a reproduction of text and figures from the following article [299]:  Windt, J, Gabbett TJ. “Is It All for Naught? What Does Mathematical Coupling Mean for Acute:Chronic Workload Ratios?” British Journal of Sports Medicine, May 28, 2018, bjsports-2017-098925.    7.2 Introduction Traditional calculations of the acute:chronic workload ratio (ACWR) are ‘mathematically coupled’, as the most recent week is included both the acute and chronic workload estimates (Figure 7.1). As Lolli and colleagues rightly point out, this induces a spurious correlation between the acute and chronic loads of ~ 0.50 (r=0.52 in their simulated data of 1000 athletes) [293]. They suggest that the simplest solution is to use uncoupled ACWRs (where the acute load is not part of the chronic load) instead.  Notably, at least two studies have already used uncoupled ACWR calculations, both demonstrating that rapid workload increases are associated with higher injury risks [212,241]. To this end, Lolli and colleagues’ suggestion warrants consideration – should we use ‘uncoupled’ ACWRs instead of ‘coupled’?  We have 2 aims in this editorial: 1) to further comment on how mathematical coupling affects ACWR estimates, and 2) to encourage 120  researchers and practitioners to use a critical approach to load management, wherein ACWRs may play a part.    Figure  7.1: The distinction between traditional ‘coupled’ and ‘uncoupled’ estimates of the acute:chronic workload ratio.  The inclusion or exclusion of the most recent week in ‘chronic load’ calculations is the key distinction.   7.3 Comments on mathematical coupling and ACWRs:  1) Defining coupled and uncoupled ACWRs. We define mathematical coupling in the same manner as Lolli et al. – where a number is represented in both the numerator and denominator of a ratio, contributing to a spurious correlation. In the case of the ACWR, both coupled and 121  uncoupled calculations convey whether recent workloads are increasing or decreasing compared to prior workloads (ACWR >1 = increasing, ACWR < 1 decreasing). However, including the most recent week in the chronic loading window changes the interpretation/definition of the ACWR.   Coupled Uncoupled Formula 𝐴0.25 ∗ (𝐴 + 𝑊2 + 𝑊3 + 𝑊4) 𝐴0.333 ∗ (𝑊2 + 𝑊3 + 𝑊4) Definition The ratio between the most recent week of work with the average of the most recent 4 weeks. The ratio of the most recent week of work with the average of the 3 preceding weeks. *Where ‘A’ is the acute workload and the last four weeks are represented by A, W2, W3, and W4, respectively.   Another way to think of the traditional, coupled ACWR is ‘what proportion of the total 4-week load is provided by the acute load?’ At constant loading conditions, the acute load makes up ¼ of the total 4-week load. As workloads spike, the acute workload constitutes a greater proportion (Table 7.1). The ACWR equals how many times greater this proportion is than 0.25. This number can approach, but never reach 4, since the acute load can constitute virtually all of the chronic load, but never all of it.  122  Table 7.1: Coupled and uncoupled (rolling averages over 4 weeks) ACWR calculations Week 1 Week 2 Week 3  Week 4 (Acute) Chronic  (with Acute) Chronic (without Acute) Traditional (Coupled) ACWR Uncoupled ACWR Total 4 Week Load Proportion of 4 Week Loads made up by Acute How many times greater is the acute proportion than 0.25? 10 10 10 0.01 7.5 10 0.00 0.001 30.01 0.000 0.00 10 10 10 0.75 7.7 10 0.10 0.075 30.75 0.024 0.10 10 10 10 2 8.0 10 0.25 0.2 32 0.063 0.25 10 10 10 4.3 8.6 10 0.50 0.43 34.3 0.125 0.50 10 10 10 7.5 9.4 10 0.80 0.75 37.5 0.200 0.80 10 10 10 10 10.0 10 1.00 1.00 40 0.250 1.00 10 10 10 14.5 11.1 10 1.30 1.45 44.5 0.326 1.30 10 10 10 18 12.0 10 1.50 1.80 48 0.375 1.50 10 10 10 30 15.0 10 2.00 3.00 60 0.500 2.00 10 10 10 50 20.0 10 2.50 5.00 80 0.625 2.50 10 10 10 90 30.0 10 3.00 9.00 120 0.750 3.00 10 10 10 210 60.0 10 3.50 21.00 240 0.875 3.50 10 10 10 100000 25007.5 10 4.00 10000 100030 1.000 4.00   123   2) The coupled correlation is expected. The ‘spurious’ correlation of approximately r = 0.5 between acute and chronic loads is exactly what we would expect based on the ACWR calculation. Squaring this correlation (r2 = 0.52 = 0.25) demonstrates that, on average, ~25% of the total variation in our outcome (chronic loads) is explained by its linear relationship with the acute load. This aligns with the fact that the acute load makes up one fourth of the chronic load calculation. Extending this logic, if a coupled ACWR was calculated over a 5-week window, the correlation would approximate r=0.45 (r2 = 0.20).  3) One can convert between coupled and uncoupled ACWRs. Since both ratios discussed by Lolli et al. are calculated from the same four training weeks, one can convert between the two via the following formula (Appendix C). *𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 𝐴𝐶𝑊𝑅 =  3 ∗ ′𝐶𝑜𝑢𝑝𝑙𝑒𝑑′𝐴𝐶𝑊𝑅(4−′𝐶𝑜𝑢𝑝𝑙𝑒𝑑′𝐴𝐶𝑊𝑅)   4) Effects of mathematical coupling on ACWRs. Lolli et al. note that a ‘coupled’ ACWR of 1.45 equals an ‘uncoupled’ ACWR of 1.71. Thus, mathematical coupling ‘suppresses’ traditional ACWRs compared with uncoupled ACWRs. The degree to which the ratio is suppressed increases as the ACWR increases. The ratios are equal at 1.0, vary slightly at first (1.5 vs. 1.8), and progressively diverge (2.0 vs. 3.0, 3.0 vs. 9.0). The coupled ACWR never exceeds 4, regardless of how high the uncoupled ACWR gets (Figure 7.2) so a coupled ACWR should not be considered a continuous measure. The way that these scaling properties affect research estimates and the appropriateness of including them in different analyses (e.g. regression modelling) deserves further consideration.  124   Figure 7.2: The association between coupled and uncoupled estimates of the acute:chronic workload ratio with increasing workload progression.  While the coupled ACWR asymptotes at 4, uncoupled estimates continue to increase in a linear fashion.  125  7.4 What now? A call for critical, thoughtful monitoring Mathematical coupling may support uncoupled ACWRs in the future. However, rather than suggest one particular approach, we propose that researchers and practitioners should focus on a thoughtful, critical approach to load monitoring, which may, or may not include ACWRs. We highlight 3 practical applications below:  1) Don’t use any ACWR in isolation  Athlete monitoring is a multifaceted process with information gleaned from external and internal loads, athlete wellness, readiness, and other information [100]. One factor which may be of interest to practitioners and researchers is the rate of workload change. An ACWR (coupled or uncoupled, rolling average or exponentially weighted) provides a single measure that indicates how recent loads are changing relative to previous loads. Whatever method (coupled or uncoupled) is chosen, it must not be interpreted in isolation. Instead, it must be interpreted in conjunction with other information – e.g. training modalities, absolute workload changes, how the athlete is tolerating training, the athlete’s chronic load, and the athlete’s internal risk profile (e.g. age, aerobic fitness level, strength, etc.) [5, 6].   2) Understand the implications of how any given ACWR was calculated.  Different acute and chronic time windows [236], exponentially weighted moving averages [294], and week-to-week changes [300] have been proposed in the literature to monitor training progression. Each of these calculations will result in different numbers given the same workload sequence. Notable ‘thresholds’ will also differ depending on how an ACWR is calculated. The proverbial ‘sweet spot’ of 0.8 to 1.3 for the ACWR has been largely based on coupled, rolling average estimates. If uncoupled (1-week:4-week) ACWR estimates were used, 126  these same thresholds would correspond to different ratios (0.75 to 1.45). It is up to the user to critically select which variation, if any, they will use, and how they will interpret the values.   3) Explicitly detail how the ACWRs were calculated.  Finally, if researchers choose to use ACWRs, they must explicitly detail how they were calculated, facilitating interpretation and replication. These details should include whether calculations were rolling averages or exponentially weighted [294], the length of the acute:chronic windows, and whether acute and chronic workloads were coupled or not.    Ultimately, developing robust, high-performing athletes depends on building high chronic workloads, but increasing these workloads too quickly will likely increase athletes’ risks of injury [74,241]. In the context of a thoughtful, critical athlete monitoring approach, ACWRs may help by estimating workload progression – however they are quantified. Chapter 8: A training load--injury paradox: Is greater pre-season participation associated with lower in-season injury risk in elite rugby league players? (Paper 6)  8.1 Preamble In the chapters 3-7, I outlined how workloads conceptually relate to injury, called researchers to thoughtfully consider what workload metrics are used, and reviewed the statistical approaches that best analyse intensive longitudinal data. In this final data chapter, I investigated whether rugby league players who participated in a greater number of pre-season sessions were better able to resist in-season injury. As detailed in Chapter 5, this research question deals with between- and within-athlete differences. The primary research question – whether completing more pre-season sessions is associated with a greater ability to stay healthy during the season – differentiated between athletes, as each athlete completes a distinct number of pre-season sessions. I also needed to control for the within-athlete exposures to workloads throughout the competitive season. Multilevel modeling allowed me to cluster weeks (level 1) within each athlete (level 2), so that weekly workload variables could be included in the model, along with athlete characteristics.   The rest of this chapter is a reproduction of text and figures from the following article [173]:  Windt, Johann, Tim J. Gabbett, Daniel Ferris, and Karim M. Khan. “Training Load--Injury Paradox: Is Greater Pre-season Participation Associated with Lower in-Season Injury Risk in Elite Rugby League Players?” British Journal of Sports Medicine, April 13, 2016, bjsports-2016-095973.  128   8.2 Introduction Athletic injuries are common in team sports [132,133], compromising team success [52,53,134] and posing a significant financial burden to organisations [54]. High training loads and substantial spikes (rapid increases) in training volume have been associated with increased injury rates [52,113,115]. Both external (work completed) [136,138,160] and internal (physiological response such as perceived exertion or heart rate) [157,159,210,301] training load measures have been used to identify the association between workloads and injury risk.  Traditionally, workload-injury investigations focused on absolute workloads and injury [142,157], and higher workloads were associated with greater rates of injuries [142]. However, high training loads are necessary for beneficial physiological adaptation such as increased aerobic capacity, strength, and repeat-sprint ability, along with optimal body composition [163,164], many of which are associated with decreased injury risks [160,166,167].   Recently, load-injury investigations have highlighted that the relationship between acute (1 week) and chronic (rolling 4-week total averaged to one week) workloads, termed the acute:chronic workload ratio, may better predict injury risk than total workloads [52,113,136]. Moreover, Hulin and colleagues demonstrated that as long as players' acute:chronic workload ratios were kept within a moderate level (0.85-1.35), high chronic workloads may reduce injury risk in rugby league players – the training load-injury paradox [112,113].   129  Pre-season training provides several physical benefits for sporting teams. It allows players to reach high chronic workloads [113], as well as to develop the physical capacities associated with reduced injury risks [160,166,167]. Indeed, pre-season periods often include higher training loads than in-season periods [52,159]. Theoretically, players who have a more ‘successful’ pre-season may be more resilient to injury when faced with the demands of the competitive season.    To our knowledge, no study has investigated whether pre-season training provides a foundation which decreases in-season injury risk in elite team sport athletes. Therefore, we investigated whether elite rugby league players who participated in a greater number of pre-season sessions were less likely to miss games due to injury throughout the competitive season, while accounting for their external training loads during the competitive season.   8.3 Methods 8.3.1 Study design We prospectively followed thirty rugby league players (mean ± SD age, 25 ± 3 years) from one elite rugby league club throughout their 17-week pre-season period and 26-round competitive season. All participants provided informed written consent and received a clear explanation of the study. All experimental procedures were approved by the Institutional Review Board for Human Investigation at Australian Catholic University.  8.3.2 Measures We collected both time-varying and time-invariant variables. Time varying variables were summarised weekly and included injury status and daily training load variables. Time invariant 130  variables included the number of pre-season sessions completed, player position, and age at the start of the pre-season. The competitive season was divided into 3 time periods for descriptive purposes, around the representative (i.e. ‘State of Origin’ interstate series) period: ‘pre-origin’ (Weeks 1-8), ‘origin’ (weeks 9-17), and ‘post-origin’ (weeks 18-26). Injury status  The team’s medical staff (including physician and physiotherapist) diagnosed injuries, while the team physiotherapist updated and maintained the injury reports. For the purpose of this investigation, an injury was defined as any injury that resulted in a loss of match-time – “match time loss only” [18]. Pre-season attendance: For each training day throughout the pre-season and competitive season, players’ participation in training was recorded as ‘full’, ‘modified’, ‘rehab’, or ‘away’. Players’ individual pre-season participation levels were quantified as the number of “full” pre-season sessions they completed. Quantifying in-season training loads External workloads were obtained using global positioning system (GPS) devices (GPSports, SPI-HPU 5 Hz (interpolated 15 Hz), Canberra, Australia). Load variables collected included total distance, high speed (>5 m/s) distance covered, and the percentage of total distance completed at high-speeds. Our analysis included all field training sessions and National Rugby League matches throughout the 2015 season.   131  8.3.3 Data collection and analysis procedures Data were categorised into weekly blocks from Monday-Sunday throughout the 26-week season. If GPS data were missing for players who were recorded as attending the ‘full’ training session, load data was estimated by calculating the average workload for players of the same position who participated in the full session. Since models were fitted to determine the likelihood of sustaining a time-loss injury in a given week or subsequent week, players’ data for a given week were excluded if they were already injured, suspended, or released from the team.   8.3.4 Statistical analysis All data were analysed in the open-source statistical software, R (Version 3.2.2). Separate univariate random effect logistic regression models were fitted for each time-varying and time-invariant variable, with the likelihood of sustaining a time loss injury as the outcome variable, and random-intercepts for each player. These models were used to determine which variables were associated with an increased or decreased risk for injury throughout the season, not controlling for other covariates. Random effect models were chosen for their ability to handle unbalanced data with varying number of follow-up observations, their capacity to generate individual specific-predictions, and for their recommended use in analysing repeated-measures designs with correlated data.  In fitting the regression models, all training load variables were standardised, due to the different scales of the measures and subsequent failure of the models to converge in the statistical software with unadjusted predictor variables. Odds ratios were calculated to determine the effect size associated with a one standard deviation increase in training load variables. For pre-season 132  participation, odds ratios were calculated to examine the effect sizes associated with an increase of 10 “full” pre-season sessions. Statistical significance was set at p<0.05 for all analyses, and odds ratios were calculated as an effect size for all models.   Two separate multilevel logistic regression models were fit to determine the effect of pre-season participation on injury likelihood, controlling for training loads. One model was fit to determine the likelihood of injury in the current week. A second model was fit to determine the likelihood of injury in the subsequent week. The final models were first fitted by including variables shown to be significant predictors from univariate models. From here, all other training load variables, as well as time invariant covariates (age, position, season period), and interaction terms were added to the model to optimise model fit. Model fit was assessed by minimising the values of model diagnostics criteria (AIC/BIC) and the standard deviation of the random intercepts. Variables that did not improve the model fit were excluded from the final models.   8.4 Results  8.4.1 Injuries A total of 40 injuries were sustained during the competitive season (29.0/1000 h). These led to 241 total matches missed. There were no significant differences in injury likelihood when comparing positions (p = 0.73) or season period (p = 0.46).   133  8.4.2 Pre-season participation During the pre-season period (November 3, 2014 to February 27, 2015), the team had 87 pre-season training sessions. Players completed an average of 64 ± 19 ‘full’ pre-season sessions (range 12 - 86).   8.4.3 Pre-season loads and injury risk Over the course of the season, there was a significant correlation between the number of full pre-season training sessions that players completed and the number of full in-season sessions completed (r = 0.59, p<0.001). Further, there was a significant association (r = -0.40, p < 0.05) between the number of pre-season sessions players completed and the percentage of games they missed due to injury (Figure 8.1). Without adjusting for training loads, greater pre-season participation was associated with a decreased likelihood of injury throughout the competitive season during the current (OR = 0.82, 95% CI = 0.69-0.97) and subsequent week (OR = 0.80, 95% CI = 0.68-0.94).   134    Figure 8.1: The association of pre-season participation with games missed due to injury.  r = -0.40, p < 0.05. Percentage of games missed due to injury was calculated by dividing the number of games missed due to injury by the number of games that players were eligible to play. Games were excluded from calculation if players were ineligible due to suspension or being traded during the season and therefore not eligible to play.    135  8.4.4 Training loads and injury risk Training load measures over the course of the competitive season are summarised in Table 8.1. The average distance as well as injury incidence for each week of the competitive season are displayed in Figure 8.2.    Table 8.1. Descriptive statistics for players’ average workload and injuries over the duration of the study.  Variable Phase of season Total  Pre-Origin Origin Post Origin  Injuries (n) 14 15 11 40 Distance (m) 9574 (4099) 10815 (5311) 8960 (5871) 9780 (5205) High Speed Distance (m) 677 (462) 668 (467) 525 (442) 621 (461) High Speed Distance/Total Distance (%) 7.2 (4.8) 6.1 (2.4) 6.0 (3.1) 6.4 (3.7) Data are mean (SD) for load measures and count for injuries.  Phase of season divides the season into three periods around the ‘State of Origin’ interstate series: ‘Pre-origin’ (Weeks 1-8), ‘origin’ (weeks 9-17), and ‘post-origin’ (weeks 18-26).   Figure 8.2: Weekly training loads and injury incidence  Higher acute (1 week) distances were associated with lower injury likelihoods in the current week (OR = 0.64, 95% CI = 0.46-0.90), but not in the subsequent week. Conversely, a greater percentage of total distance completed at high speeds was associated with an increased likelihood of injury both in the current (OR = 1.34, 95% CI = 1.03-1.73) and subsequent week (OR = 1.07, 95% CI = 1.06-1.08). Absolute high-speed running distance was not associated with injury likelihood in either the current or subsequent week.   Chronic workloads were not significantly associated with injury risk in either the current or subsequent weeks. Similarly, acute:chronic workload ratios for all training load variables were not associated with significant changes in injury likelihood. Table 8.2 summarises all models of single training load variables and their associated effects on injury risk.    138  Table 8.2 Association of training load variables with injury likelihood in the current and subsequent week.  Load calculation  1 SD Effect of 1 SD increase on current week injury likelihood  OR (95% CI) Effect of 1 SD increase on subsequent week injury likelihood  OR (95% CI) Acute (1 Week Loads) Distance High Speed Distance High Speed Distance/Total Distance (%)   5205 m 461 m 3.7%  0.64* (0.46-0.90) 0.83 (0.58-1.20) 1.34* (1.03-1.73)  0.86 (0.61-1.22) 0.83 (0.57-1.19) 1.07* (1.06-1.08) Chronic (Rolling 4 Week Average Loads) Distance High Speed Distance High Speed Distance/Total Distance (%)   3389 m 346 m  3.1%   0.80 (0.58-1.11) 0.86 (0.60-1.22) 1.13 (0.83-1.54)   0.82 (0.57-1.18) 0.99 (0.70-1.40) 1.22 (0.90-1.66) Acute:Chronic Workload Ratios Distance High Speed Distance High Speed Distance/Total Distance (%)  0.72 0.90 0.42  0.72 (0.48-1.07) 0.91 (0.64-1.31) 1.64 (0.83-3.24)  0.89 (0.63-1.27) 0.88 (0.61-1.27) 1.13 (0.53-2.42) * p < 0.05, SD = Standard Deviation   139  8.4.5 Full injury prediction models Two multivariate injury prediction models quantified the effect of pre-season participation on injury risk, controlling for training loads (Table 8.3). Training load variables included in the final models were those that were found to have a significant association with injury risk in univariate models, specifically 1-week total distance and the 1-week proportion of distance performed at high speeds. The fit of these final models was not improved with the addition of any other variable, nor with the addition of a random slope to the model, so none were included.   Table 8.3. Effect of pre-season participation on injury likelihood in current and subsequent week, controlling for training load variables. Model Variable OR (95% CI) Model 1: Injury Likelihood in Current Week 10 Pre-Season Sessions High Speed Distance/Total Distance Acute Distance (1 SD increase) 0.85 (0.70-1.02) 1.27 (0.99-1.63) 0.56* (0.36-0.87) Model 2: Injury Likelihood in Subsequent Week 10 Pre-Season Sessions  High Speed Distance/Total Distance Acute Distance 0.83* (0.70-0.99) 1.07 (0.79-1.45) 0.82 (0.55-1.21) * p<0.05  Model 1 predicts the likelihood of injury in the current week. Controlling for training load, increased pre-season participation was still associated with a reduced odds of injury, though this was no longer statistically significant (OR = 0.85, 95% CI = 0.70-1.02). Similarly, a greater percentage of distance ran at high speeds appeared to be associated with an increase injury risk, but the effect was no longer significant when controlling for pre-season participation and acute distance (OR = 1.27, 95% CI = 0.99-1.63). Finally, as with univariate models, greater acute distance was associated with a significantly reduced likelihood of injury (OR = 0.56, 95% CI = 0.36-0.87).  140   Model 2 predicts the likelihood of injury in the subsequent week with a given pre-season participation, acute distance, and acute percentage of distance ran at high speeds. In this model, when controlling for distance and percentage of distance at high speed, increased pre-season participation was associated with a reduced likelihood of injury (OR= 0.83, 95% CI = 0.70-0.99). Within this model, neither acute distance nor percentage of distance ran at high speeds were significantly associated with injury risk in the subsequent week (Figure 8.3).    141     Figure 8.3: Predicted injury probabilities based on high speed running percentage and pre-season participation. Injury probabilities were predicted from model 2, predicting the probability that a player would sustain a match time-loss injury in a subsequent week, controlling for the total distance and percentage of total distance ran at high speeds in the current week, as well as pre-season participation. Pre-season participation has been divided into three equal tertiles, such that “Low” equals < 59 sessions (n=10), “Moderate” equals 59 – 75 sessions (n=10), and “High” equals >75 sessions (n=10).   142  8.5 Discussion In this sample of rugby league players, greater pre-season participation was associated with a decreased injury risk during the competitive season. Players who participated in a greater number of full pre-season sessions had a reduced likelihood of injury throughout the competitive season, completed more in-season training sessions and missed fewer games due to injury. This reduced injury likelihood in the subsequent week was maintained even when controlling for training load variables (acute distance and acute percentage of total distance completed at high speeds).   8.5.1 Training loads and injury likelihood In contrast to recent studies [52,113], neither acute:chronic workload ratios nor chronic workloads in isolation were significantly associated with injury. It may be that the current sample size was too small to detect these effects. However, two separate measures of acute (1 week) training loads, 1) distance covered, and 2) percentage of total distance covered at high speeds, were significantly associated with changes in injury likelihood.   Higher 1-week total distances were associated with a reduction in injury likelihood in the current week, but not in the subsequent week. Similarly, previous data in rugby league players showed that greater distances completed at lower intensities were associated with a reduction in injury risk [160]. However, it should be noted that the reduced likelihood of injury associated with higher distances in the current week may be partly attributable to players sustaining an injury earlier in the week. In this case, it may be that increased distances are not preventing injury, but are accumulated by players who are healthy.   143  In contrast to total distance, the percentage of total weekly distance performed at high speeds was associated with an increased risk of injury in the current and subsequent week. This increased injury risk with greater high-speed running loads has been previously seen in team sport athletes [138,160]. Notably, when controlling for players’ pre-season attendance, the percentage of running performed at high-speeds was no longer significantly associated with injury. This may indicate that players who had accumulated more of the benefits of a successful pre-season were better able to tolerate the stress of the competitive season. Similarly, it has previously been shown that team sport athletes who perform greater than 18 weeks of training before sustaining initial injuries were at a reduced risk of sustaining a subsequent injury [302]. Collectively, the present and previous [23] results demonstrate the protective effect of pre-season training in team sport athletes.  8.5.2 Pre-season participation and in-season injury – providing protection or revealing underlying differences [i.e. identifying the robust players]? We speculate as to potential mechanisms responsible for the associated reduction in injury likelihood with increased pre-season participation. From a physical standpoint, pre-season participation may be protective by allowing players to accumulate high chronic workloads [113], and develop greater strength and aerobic capacity [302]. Further, players who participate in a greater proportion of pre-season training sessions may also be better prepared mentally and tactically within the team environment. On the other hand, increased pre-season participation may merely identify players who are inherently more robust to injury, and therefore more likely to handle both the pre-season training loads as well as the rigors of the competitive season. Thus, the association between pre-season training and in-season injury risk may stem from both protective and revelatory effects of the pre-season.  144   8.5.3 Preparation through pre-season participation – practical implications Our findings do not necessarily indicate a need for high training loads during the pre-season. Rather, pre-season training should be conducted so that players are able to participate in the highest proportion of team sessions possible. This is especially pertinent given that the majority of training-related rugby league injuries occur during the pre-season periods [158]. Even though 24% of a rugby  union team’s annual training occurred during the pre-season period, 34% of training related injuries occur during this time [303]. High injury incidence during the pre-season period may be partly attributable to training loads, which are generally higher in pre-season periods compared with in-season periods [52,302]. Moreover, a reduction in pre-season training loads significantly reduced injury incidence in rugby league players [158]. From our experience (TG), those who manage team training loads should aim to design pre-season training periods which induce positive physiological adaptations while minimising injury risk and maximising player availability. Accurate season-by-season records of injury and training loads will help teams find their ‘sweet spot’.   8.5.4 Potential limitations The present study included a sample size (30 players), and total injury occurrences (40), which limits the number of variables that could be included in the injury prediction models and reduces the sensitivity of the models. Due to the limited availability of GPS devices, the number of players who could be monitored was restricted to 30, as opposed to all members of the rugby league club. Although more injuries would have been captured with a broader definition of injury, “match time-loss only” is accepted as an accurate and reliable definition used in team sport contexts [18].  145   Although we used GPS-derived total and high speed running distances, the inclusion of other external load variables (e.g. accelerations, decelerations, collisions) would likely add value to any investigation of the relationship between pre-season training load and in-season injury risk in rugby league players [96,101]. However, as discussed in recent load injury investigations [113,168], the ability of GPSports technology to accurately measure these variables is limited. Further, internal load measures (e.g. session-RPE or heart-rate) may also be useful to investigate pre-season and in-season training loads. Finally, while we quantified pre-season participation in sessions, future investigations may quantify pre-season training loads to further distinguish characteristics of pre-season training that facilitate the development of athlete resilience.  8.5.5 Summary and conclusions In this first study to investigate the association of pre-season training participation and injury likelihood during the competitive season, players who completed a greater number of pre-season sessions were less likely to be injured during the competitive season, even when controlling for their external training loads. Total distance covered was associated with a decreased likelihood of injury in the current week, while players who completed a greater percentage of their total distance at high-speeds were at increased risk of injury in the current and subsequent week.  146  Chapter 9: Conclusion 9.1 Integrating my findings into the field, and contributions of my dissertation  Workload and injury research has proliferated in the last decade. From the time I commenced this dissertation in 2015 until today, the field has developed substantially. As just one example, the acute:chronic workload ratio was first introduced by this name in 2014. Just four years later, approximately 30 original cohort studies have investigated workloads and injury using this derived measure. Discussions have progressed to nuances of how to best calculate workload progression [293,294], and differentiating its association with injury from its predictive power [304,305]. Therefore, integrating this dissertation within the current research landscape is analogous to parking on a rapidly moving freeway.   9.1.1 Conceptual contribution My extension of the sports injury aetiology literature – the workload—injury aetiology model [131] (Chapter 3, Figure 3.6) – provided the first athletic injury aetiology model that explicitly incorporated training and competition workloads. It provided researchers and applied sports medicine/science practitioners with a framework to understand how workloads may contribute to injury occurrence, specifically through 1) exposure to potential inciting events, 2) positive physiological adaptations which improve athletes’ individual risk profiles, and 3) negative fatiguing effects on an athletes’ individual risk profiles.   Although this workload—injury aetiology model was the first to incorporate workloads explicitly into athletic injury development, at least three other frameworks have since extended the aetiological literature [180,306,307]. These extensions included 1) differentiating physiological 147  from biomechanical adaptations [306], 2) expanding intra-session injury aetiology [307], and 3) introducing complex systems thinking [180].   Differentiating physiological and biomechanical adaptation timelines extends the overall workload—injury aetiology model. Vanrenterghem and colleagues introduced a similar ‘fitness’ and ‘fatigue’ athlete monitoring framework aimed at team sports, but distinguished ‘physiological’ from ‘biomechanical’ adaptations [306]. The former refers to adaptation from a central, cardiovascular perspective (e.g. lactate threshold, oxygen consumption) and the latter to the musculoskeletal system (e.g. bone, tendon, joint). Thus, a given workload progression may allow sufficient adaptation from a cardiovascular perspective, but exceed the ability of the biomechanical system to adapt following training stimuli. Where the workload—injury aetiology model focused on injury development as the outcome, Vanrenterghem and colleagues quantified external loads and internal loads with a spectrum of underloading to overloading as their outcome. Notably, both these models conceptualised training and adaptation over time and across multiple sessions, with less emphasis on intra-session injury aetiology.   Intrasession injury aetiology among runners was the central focus of Bertelsen and colleagues’ running injury framework [307]. Within their framework, injury occurs when the load on a specific structure exceeds the capacity of that structure. The rest of the model details the factors contributing to the structure-specific loading (divided into load magnitude and load distribution per stride) and the structure-specific capacity. Like my model, they considered internal risk factors (e.g. previous injury, sleep, age, etc.) and external risk factors (e.g. shoes, terrain). Risk factors can alter structure-specific capacity (e.g. a previous injury may have worsened tissue integrity) or 148  structure-specific loading – through the load magnitude per stride (e.g. body weight) or through the load distribution per stride (e.g. terrain, anthropometrics). On one hand, the framework’s increased granularity provides more clarity on how risk factors contribute to injury risk at a tissue level. On the other hand, there is no recursive element, so the structure-specific adaptation across sessions is not incorporated into the model.   Complex systems thinking presented a third development in the aetiological literature. Extending other’s work [225], Brazilian PT/PhD Natalia Bittencourt and colleagues introduced a complex systems approach to athletic injury aetiology that fundamentally shifted the discussion away from internal and external risk factors to a ‘web of determinants’ made up of these factors (Figure 5.1) [180]. Within this framework, athletic injury is an ‘emergent pattern’, which can occur through multiple ‘risk profiles’. Risk profiles are made up of ‘regularities’ where certain risk factors interact and contribute jointly to injury risk. Complex systems analyses are conceptually appealing yet challenging to analyse, and statistical approaches (e.g. agent-based modeling) require datasets that are larger than those currently collected in sporting settings. To address this challenge, simulated data can be used to explore injury aetiology using complex systems analyses, as demonstrated in the first of these studies [238]  Extending the workload—injury aetiology model, mediation and moderation are causal models that I suggested may help explain the workload—injury association (Chapter 4). Within the last three years, several original investigations have supported this notion [182,241,245,308]. Aerobic fitness moderated the workload—injury association in both elite Gaelic football and professional soccer [182,309], with fitter players more robust to workload spikes measured with an ACWR. 149  Heart rate variability also moderated workload spike tolerance in a small sample of Crossfit athletes over a 16-week period [245]. Finally, scapular dyskinesis moderated the relationship between throwing workload and shoulder injury in 679 elite handball players [241].  Using the handball study, Nielsen and colleagues extended the moderation discussion by drawing from their running aetiology framework [307]. They presented a directed acyclic graph (DAG) to demonstrate the potential causal pathway to shoulder injury [220]. In this DAG they again differentiated how workloads impact injury through structure-specific loads exceeding that structure’s capacity. In this case, scapular dyskinesis modifies the distribution of structure-specific loads onto certain muscles and contributes to injury at a workload where an athlete with normal scapular kinematics would be distributing the load more evenly and thus be protected against injury.    9.1.2 Methodological contributions Although several systematic reviews and an international consensus statement on workloads and injury have been published, none had critically appraised the statistical approaches that authors employed.  I reviewed and then applied US Professor Linda Collins’ threefold framework (1 – conceptual model, 2 – temporal design/data collection frequency, and 3 – statistical model) [126] as a lens to assess these statistical approaches (Chapter 5). The inclusion criteria meant that the theoretical aetiology models which described daily fluctuating injury risk were aligned with the daily data collection in all of the included prospective cohort studies, leaving the statistical model to consider.  150  Very few sport workload studies had utilised statistical approaches that can adequately handle the challenges imposed by intensive longitudinal data. One of these challenges – the dependency created by repeated measures – increases the risk of false-positive findings when analysing workload—injury data [259].   Several statistical approaches (e.g. correlation, t-tests, ANOVA) used in the past also cannot answer research questions begged by existing aetiological frameworks – such as whether risk factors moderate the workload—injury association. I recommended multilevel/mixed modeling and time-to-event modeling as well-suited approaches that have been underutilised in the field. Given the superiority of multilevel modeling over generalised estimating equations, I provided two examples of mixed/multilevel modelling studies using longitudinal data in Chapter 6 and Chapter 8. Since ILD are not limited to sports medicine, methodologically reviewing the strengths and weaknesses of analytical approaches may inform future research across disciplines.  Methodological and statistical considerations for longitudinal workload data continue to move to the forefront of research – this is a dynamic field. Therefore, other statistical approaches and methodological considerations that I did not discuss in detail within this thesis are likely to contribute to the field in the future. Simulation research has recently used complex systems approaches (specifically, agent-based modeling) to explore the workload—injury association in distance runners [238], and to demonstrate how non-linear approaches such as regression splines are more appropriate than discretization for modeling the ACWR [259].   151  9.1.3 Applied contributions Given the advances in research methodology in athlete monitoring, there is an opportunity to apply these considerations into applied practice (e.g. working with teams). In Chapter 7, I responded to published criticism of the ACWR [293] by describing how mathematical coupling influences the derived calculation of the ACWR, and what this effect means for applied sport scientists and sports medicine researchers. I concluded that researchers must explicitly report what measures they chose and think critically about the implications. The same workload progression will result in different ACWR values depending on how it is calculated, and the variable itself may change from linear, unbounded variable (uncoupled) to an upper-bounded non-linear variable (coupled). Therefore, the effect of any specific ACWR value will greatly depend on how it is derived (coupled vs. uncoupled, rolling averages vs. exponentially weighted). Similarly, applied practioners often use decision support systems or ‘traffic light systems’ to alert them to ‘notable’ findings [226,310]. Since given workload progressions give different ACWR values, practitioners must critically consider what each calculation method means for various ‘threshold’ calculations and use this to inform what findings they deem ‘notable’.   To my knowledge, Chapter 8 was the first study to investigate whether a successful pre-season – judged by the number of sessions completed – helped to prepare athletes for regular season demands. It was also the first prospective cohort study in the workload—injury field to use a mixed/multilevel model. By using a mixed modeling approach, as recommended in Chapter 5, I clustered weekly workload observations (Level 1) within each athlete (Level 2). By structuring the data in this multilevel format, athlete characteristics (e.g. age, # of preseason sessions) could be modeled together with workload, with injury as the outcome.   152   Looking through the lens of the framework in Chapter 5, this modeling approach allowed me to address two major themes from theoretical aetiology models (like the workload—injury aetiology). I considered a multifactorial injury aetiology and included both within-athlete (workloads) and between-athlete (pre-season sessions) variables. This modeling approach also addressed three longitudinal data analysis challenges – 1) it controlled for the repeated measures dependency by clustering observations (weeks) within athletes, 2) it adequately handling unbalanced data as athletes had different numbers of observations, and 3) it included time-varying (weekly workloads) and time-invariant variables (e.g. age and number of pre-season sessions) in the model.   Mixed modeling enabled me to show that athletes who completed more preseason sessions were able to complete a greater proportion of in-season sessions and were better prepared for in-season demands. These preseason benefits have since been documented in Australian football [201,311] and in American football [312]. Collectively, these studies suggest that players who participate in a greater proportion of a team’s planned pre-season may be more robust during the regular season. This could relate to a ‘survival of the fittest’ phenomenon, or to athletes successfully building up a sufficient chronic workload and adaptation during the preseason period. As these published studies examined single teams, it remains to be seen whether this extends to between-team comparisons – namely, whether teams with more pre-season sessions have lower injury rates than those with fewer pre-season sessions.  153  9.2 Dissertation strengths The first strength of this dissertation is its methodological focus. Workloads are an important causal factor related to injury risk. However, understanding how workloads relate to injury and the strength of this relationship depends on the strength of the analyses employed in the original manuscripts. Although there have been many original investigations and now a reasonable number of systematic reviews in our field, methodological appraisals of these studies had been entirely lacking. The methodological emphasis came at a time where increasing amounts of longitudinal data were being collected. My review of these original studies through an intensive longitudinal data analysis framework should inform future analyses in sports medicine and has the potential to provide insights for other health researchers who benefit from intensive longitudinal data. As one example, Prof. Collins’ lab recently used multilevel models to better understand how stress affected drinking behaviour in college students, based on daily diary data [249].    Using multilevel modeling to account for the repeated-measures nature of the data included in the original prospective cohort studies represented a step forward in the modelling procedures in many workload—injury studies. Subsequent simulations have demonstrated that failing to account for the repeated-measures nature of the data in longitudinal workload monitoring increases the false-positive rate [259].   9.3 Dissertation limitations - what I would do differently  My studies described in Chapters 6 and 8 were among the first to incorporate multilevel modelling with workload and injury data, and among relatively few that controlled for the repeated measures nature of longitudinal data. Reporting and interpreting the random effects from mixed models is 154  another benefit of mixed models. In hindsight, I would have incorporated these to a greater effect in both mixed modeling chapters. Further, more advanced statistical techniques have continued to be developed, particularly time-to-event models that account for competing risk and multi-state transitions. Moving forward I would utilise these methods in prospective cohort studies, if there were enough events per variable to power these models.   I modeled ACWR as a continuous variable in my pre-season analysis (Chapter 8), instead of through non-linear approaches. This may contribute to why ACWR was not associated with injury risk. Most of the studies that have included ACWR to injury risk, explored potential non-linear associations through discretization/categorisation of the ACWR [259] and suggest a non-linear (u-shaped) association between the ACWR and injury risk. Carey and colleagues simulated workload data and demonstrated how regression splines or non-linear modelling may be more appropriate than discretization [259]. I would revisit our modeling to incorporate ACWR in this way.   9.4 Future research directions Global (internal) workload measures (e.g. session RPE) and ‘standard’ workload progression measures (e.g. 1- to 4-week rolling average ACWRs) are unlikely to be ideal for all sporting environments and injury types. Refining which load measures and derived workload variables most closely relate to injury risk will continue. Sport-specific load measures may include jump count for sports such as volleyball [313,314], or throw counts for sports such as handball or baseball [241,315]. Researchers may also differentiate whether running (e.g. high-speed running distance) or change of direction (e.g. accelerations, decelerations) metrics are most relevant in team field/court sports [317]. 155   Ideal derived variables may also vary across sporting contexts or injury types. Standard 1-week to 4-week rolling averages may be less sensitive than ratios with different acute/chronic time-windows or exponentially weighted approaches  [236,246,317]. Since the timeline for adaptation differs based on the specific system (e.g. physiological/biomechanical), different acute:chronic ratios relate more closely to specific injury types based on tissue type. Finally, given the criticisms of rolling average ACWR approaches [290,291,293], other derived variables may be developed or refined that that more accurately capture workload progression.   By using appropriate multivariable statistical approaches, future workload—injury investigations may better understand athletic injury aetiology by asking theory-driven questions. Which risk factors moderate the workload—injury association? Do multiple risk factors interact with each other or with workload progression to alter injury risk?   Injury prediction remains the ‘holy grail’ of athletic injury research [319]. Athlete screening with single physical tests cannot be used to accurately predict injury at an individual athlete basis [183]. Existing analyses using workload data have also been relatively poor predictors to date [304,305,319]. However, these poor predictive capabilities may largely be expected because many used only one derived workload variable (e.g. ACWR) to predict injuries. Since injury risk is multifactorial, as datasets grow to include multiple risk factors along with workload data, future investigations should undoubtedly deploy more complex predictive models, the performance of which remain to be seen.   156  Injury aetiology – a cornerstone of my dissertation – is the second step of van Mechelen’s sequence of prevention model [5]. Although there is an association between workloads and injury, injury prevention can only occur when this knowledge informs preventative interventions. An important next step will be for researchers to (i) develop a preventive intervention that includes workload management (e.g. controlling the number of times athletes are exposed to rapid workload increases while building high chronic workloads to maximise positive adaptation) and (ii) test its efficacy can be evaluated in a research context (e.g. randomised control trial). I understand that such a trial has been registered in Norway.   As the TRIPP model [63] and recent views on the van Mechelen model [321] remind us, the implementation context of interventions is key. What is needed for successful intervention? When considering workload management to prevent injuries, it will be critical to understand which professionals dictate athlete loads in specific sporting environments (e.g. usually a head coach – not a health professional or a sports scientist). This has led to claims that the team coach/manager is the biggest injury risk a player faces. Future research will likely address whether their coaches’ behaviours can be modified. As I was finalising this thesis, a paper gained a lot of attention showing that the quality of communication between the coach/manager and the health professional team influenced team injury rates in professional football.   In conclusion, the workload—injury field has several avenues for future research. Treating this thesis as a hub, I see the potential for at least four spokes in the domains of: (i) sport-specific workload measures and derived variables (e.g., jumping in volleyball), (ii) aetiological investigations that incorporate workloads in addition to traditional risk factors and apply 157  appropriate longitudinal data analyses, (iii) injury prediction models that incorporate multiple variables from larger datasets as technological advancements continue, and (iv) clinical trials that employ best practice workload management.   Broadly speaking, my 2015-18 PhD experience provided me invaluable lessons in both research and life. In research, I learned that the right question makes all the difference; in life, I learned that the right team makes all the difference.    158  Bibliography 1  Khan KM, Thompson AM, Blair SN, et al. Sport and exercise as contributors to the health of nations. Lancet 2012;380:59–64. 2  Warburton DER, Nicol CW, Bredin SSD. Health benefits of physical activity: the evidence. CMAJ Can Med Assoc J 2006;174:801–9. doi:10.1503/cmaj.051351 3  Hewett TE, Myer GD, Ford KR. Anterior cruciate ligament injuries in female athletes: part 1, mechanisms and risk factors. Am J Sports Med 2006;34:299–311. doi:10.1177/0363546505284183 4  Meeuwisse WH, Tyreman H, Hagel B, et al. A dynamic model of etiology in sport injury: the recursive nature of risk and causation. Clin J Sport Med 2007;17:215–9. doi:10.1097/JSM.0b013e3180592a48 5  van Mechelen W, Hlobil H, Kemper HC. Incidence, severity, aetiology and prevention of sports injuries. A review of concepts. Sports Med 1992;14:82–99. 6  Fuller C. Injury definitions. In: Verhagen E, van Mechelen W, eds. Sports injury research. New York: Oxford University Press 2010. 43–53. 7  Langley J, Brenner R. What is an injury? Inj Prev 2004;10:69–71. doi:10.1136/ip.2003.003715 8  Brooks JHM, Fuller CW. The influence of methodological issues on the results and conclusions from epidemiological studies of sports injuries. Sports Med 2006;36:459–72. doi:10.2165/00007256-200636060-00001 9  Fuller CW, Ekstrand J, Junge A, et al. Consensus statement on injury definitions and data collection procedures in studies of football (soccer) injuries. Br J Sports Med 2006;40:193–201. doi:10.1136/bjsm.2005.025270 10  Fuller CW, Molloy MG, Bagate C, et al. Consensus statement on injury definitions and data collection procedures for studies of injuries in rugby union. Br J Sports Med 2007;41:328–31. doi:10.1136/bjsm.2006.033282 11  Junge A, Engebretsen L, Alonso JM, et al. Injury surveillance in multi-sport events-the IOC approach. Br J Sports Med 2008;42:413-421; doi: 10.1136/bjsm.2008.046631 12  King DA, Gabbett TJ, Gissane C, et al. Epidemiological studies of injuries in rugby league: suggestions for definitions, data collection and reporting methods. J Sci Med Sport 2009;12:12–19. 13  Orchard J, Newman D, Stretch R, et al. Methods for injury surveillance in international cricket. South Afr J Sports Med 2005;17:18–28. 159  14  Pluim BM, Fuller CW, Batt ME, et al. Consensus statement on epidemiological studies of medical conditions in tennis, April 2009. Br J Sports Med 2009;43:893–7. doi:10.1136/bjsm.2009.064915 15  Turner M, Fuller CW, Egan D, et al. European consensus on epidemiological studies of injuries in the thoroughbred horse racing industry. Br J Sports Med 2012;46:704–8. doi:10.1136/bjsports-2011-090312 16  Bahr R. No injuries, but plenty of pain? On the methodology for recording overuse symptoms in sports. Br J Sports Med 2009;43:966–72. doi:10.1136/bjsm.2009.066936 17  Clarsen B, Bahr R. Matching the choice of injury/illness definition to study setting, purpose and design: one size does not fit all! Br J Sports Med 2014;48:510–2. doi:10.1136/bjsports-2013-093297 18  Orchard J, Hoskins W. For debate: consensus injury definitions in team sports should focus on missed playing time. Clin J Sport Med 2007;17:192–6. doi:10.1097/JSM.0b013e3180547527 19  Clarsen B, Myklebust G, Bahr R. Development and validation of a new method for the registration of overuse injuries in sports injury epidemiology: the Oslo Sports Trauma Research Centre (OSTRC) Overuse Injury Questionnaire. Br J Sports Med 2013;47:495–502. doi:10.1136/bjsports-2012-091524 20  Clarsen B, Rønsen O, Myklebust G, et al. The Oslo Sports Trauma Research Center questionnaire on health problems: a new approach to prospective monitoring of illness and injury in elite athletes. Br J Sports Med 2014;48:754–60. doi:10.1136/bjsports-2012-092087 21  Statistics (US) NC for H. The International Classification of Diseases: 9th Revision, Clinical Modification: ICD-9-CM. 1991.  22  National Centre for Classification in Health (Australia). The International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Australian Modification (ICD-10-AM). National Centre for Classification in Health. 2004.  23  American Medical Association. Physicians’ current procedural terminology: CPT. The Association 1989.  24  Finch CF, Boufous S. Do inadequacies in ICD-10-AM activity coded data lead to underestimates of the population frequency of sports/leisure injuries? Inj Prev 2008;14:202–204. 25  Rae K, Britt H, Orchard J, et al. Classifying sports medicine diagnoses: a comparison of the International classification of diseases 10-Australian modification (ICD-10-AM) and the Orchard sports injury classification system (OSICS-8). Br J Sports Med 2005;39:907–911. 160  26  Meeuwisse WH, Wiley JP. The sport medicine diagnostic coding system. Clin J Sport Med 2007;17:205–7. doi:10.1097/JSM.0b013e318057518f 27  Finch CF, Orchard JW, Twomey DM, et al. Coding OSICS sports injury diagnoses in epidemiological studies: does the background of the coder matter? Br J Sports Med 2014;48:552–6. doi:10.1136/bjsports-2012-091219 28  Orchard J, Rae K, Brooks J, et al. Revision, uptake and coding issues related to the open access Orchard Sports Injury Classification System (OSICS) versions 8, 9 and 10.1. Open Access J Sports Med 2010;1:207–14. doi:10.2147/OAJSM.S7715 29  Rae K, Orchard J. The Orchard sports injury classification system (osics) version 10. Clin J Sport Med 2007;17:201–4. doi:10.1097/JSM.0b013e318059b536 30  McManus A. Validation of an instrument for injury data collection in rugby union. Br J Sports Med 2000;34:342–7. doi:10.1136/bjsm.34.5.342 31  Barden C, Stokes K. Epidemiology of injury in elite english schoolboy rugby union: a 3-year study comparing different competitions. J Athl Train 2018;53:514–20. doi:10.4085/1062-6050-311-16 32  Andersen TE, Floerenes TW, Arnason A, et al. Video analysis of the mechanisms for ankle injuries in football. Am J Sports Med 2004;32:69–79. doi:10.1177/0363546503262023 33  McKay CD, Tufts RJ, Shaffer B, et al. The epidemiology of professional ice hockey injuries: a prospective report of six NHL seasons. Br J Sports Med 2014;48:57–62. doi:10.1136/bjsports-2013-092860 34  Brukner P, Khan K. Sports injuries: Acute. In: Brukner and Khan’s Clinical Sports Medicine. (4th edition) Australia: McGraw Hill 2012. 15–24. 35  Brukner P, Khan K. Sports injuries: Overuse. In: Brukner and Khan’s Clinical Sports Medicine. (4th edition) Australia: McGraw Hill 2012. 25–40. 36  Bahr R, Krosshaug T. Understanding injury mechanisms: a key component of preventing injuries in sport. Br J Sports Med 2005;39:324–9. doi:10.1136/bjsm.2005.018341 37  Fuller CW, Bahr R, Dick RW, et al. A framework for recording recurrences, reinjuries, and exacerbations in injury surveillance: Clin J Sport Med 2007;17:197–200. doi:10.1097/JSM.0b013e3180471b89 38  Finch CF, Cook J. Categorising sports injuries in epidemiological studies: the subsequent injury categorisation (SIC) model to address multiple, recurrent and exacerbation of injuries. Br J Sports Med 2014;48:1276–80. doi:10.1136/bjsports-2012-091729 161  39  Hamilton GM, Meeuwisse WH, Emery CA, et al. Past injury as a risk factor: an illustrative example where appearances are deceiving. Am J Epidemiol 2011;173:941–8. doi:10.1093/aje/kwq461 40  Shrier I, Steele RJ, Zhao M, et al. A multistate framework for the analysis of subsequent injury in sport (M-FASIS). Scand J Med Sci Sports 2016;26:128–39. doi:10.1111/sms.12493 41  Shrier I, Clarsen B, Verhagen E, et al. Improving the accuracy of sports medicine surveillance: when is a subsequent event a new injury? Br J Sports Med 2017;51:26-28. doi:10.1136/bjsports-2016-096160 42  Shrier I, Zhao M, Piché A, et al. A higher sport-related reinjury risk does not mean inadequate rehabilitation: the methodological challenge of choosing the correct comparison group. Br J Sports Med 2017;51:630–5. doi:10.1136/bjsports-2016-096922 43  Toohey LA, Drew MK, Fortington LV, et al. An updated subsequent injury categorisation model (sic-2.0): data-driven categorisation of subsequent injuries in sport. Sports Med 2018;48:2199-2210. doi:10.1007/s40279-018-0879-3 44  Ullah S, Gabbett TJ, Finch CF. Statistical modelling for recurrent events: an application to sports injuries. Br J Sports Med 2014;48:1287–93. doi:10.1136/bjsports-2011-090803 45  Emery CA. Injury prevention in kids’ adventure and extreme sports: future directions. Res Sports Med 2018;26:199–211. doi:10.1080/15438627.2018.1452239 46  Parachute. The cost of injury in Canada. Toronto, ON: Parachute 2015.  47  Billette J-M, Janz T. Injuries in Canada: Insights from the Canadian Community Health Survey. 2015. (accessed 15 Jul 2018). 48  Finch C, Cassell E. The public health impact of injury during sport and active recreation. J Sci Med Sport 2006;9:490–7. doi:10.1016/j.jsams.2006.03.002 49  Pickett W, Molcho M, Simpson K, et al. Cross national study of injury and social determinants in adolescents. Inj Prev 2005;11:213–8. doi:10.1136/ip.2004.007021 50  Finch CF. Getting sports injury prevention on to public health agendas – addressing the shortfalls in current information sources. Br J Sports Med 2012;46:70–4. doi:10.1136/bjsports-2011-090329 51  Arnason A, Sigurdsson SB, Gudmundsson A, et al. Physical fitness, injuries, and team performance in soccer. Med Sci Sports Exerc 2004;36:278–85. doi:10.1249/01.MSS.0000113478.92945.CA 162  52  Cross MJ, Williams S, Trewartha G, et al. The influence of in-season training loads on injury risk in professional rugby union. Int J Sports Physiol Perform 2016;11:350-5. doi:10.1123/ijspp.2015-0187 53  Eirale C, Tol JL, Farooq A, et al. Low injury rate strongly correlates with team success in Qatari professional football. Br J Sports Med 2013;47:807–8. doi:10.1136/bjsports-2012-091040 54  Hickey J, Shield AJ, Williams MD, et al. The financial cost of hamstring strain injuries in the Australian Football League. Br J Sports Med 2014;48:729–30. doi:10.1136/bjsports-2013-092884 55  Ekstrand J. Keeping your top players on the pitch: the key to football medicine at a professional level. Br J Sports Med 2013;47:723–4. doi:10.1136/bjsports-2013-092771 56  Drew MK, Raysmith BP, Charlton PC. Injuries impair the chance of successful performance by sportspeople: a systematic review. Br J Sports Med 2017;51:1209-14. doi:10.1136/bjsports-2016-096731 57  Raysmith BP, Drew MK. Performance success or failure is influenced by weeks lost to injury and illness in elite Australian track and field athletes: A 5-year prospective study. J Sci Med Sport 2016;19:778–83. doi:10.1016/j.jsams.2015.12.515 58  Hägglund M, Waldén M, Ekstrand J. Previous injury as a risk factor for injury in elite football: a prospective study over two consecutive seasons. Br J Sports Med 2006;40:767–72. doi:10.1136/bjsm.2006.026609 59  Barber-Westin SD, Noyes FR, Smith ST, et al. Reducing the risk of noncontact anterior cruciate ligament injuries in the female athlete. Phys Sportsmed 2009;37:49–61. doi:10.3810/psm.2009.10.1729 60  Khan T, Alvand A, Prieto-Alhambra D, et al. ACL and meniscal injuries increase the risk of primary total knee replacement for osteoarthritis: a matched case–control study using the Clinical Practice Research Datalink (CPRD). Br J Sports Med 2018. doi:10.1136/bjsports-2017-097762 61  Lohmander LS, Englund PM, Dahl LL, et al. The long-term consequence of anterior cruciate ligament and meniscus injuries. Am J Sports Med 2007;35:1756–69. doi:10.1177/0363546507307396 62  Cottler LB, Abdallah AB, Cummings SM, et al. Injury, pain, and prescription opioid use among former National Football League (NFL) players. Drug Alcohol Depend 2011;116:188–94. doi:10.1016/j.drugalcdep.2010.12.003 63  Finch C. A new framework for research leading to sports injury prevention. J Sci Med Sport 2006;9:3–9. doi:10.1016/j.jsams.2006.02.009 163  64  Nielsen RO, Debes-Kristensen K, Hulme A, et al. Are prevalence measures better than incidence measures in sports injury research? Br J Sports Med 2017. Online First. doi:10.1136/bjsports-2017-098205 65  Soligard T, Schwellnus M, Alonso J-M, et al. How much is too much? (Part 1) International Olympic Committee consensus statement on load in sport and risk of injury. Br J Sports Med 2016;50:1030–41. doi:10.1136/bjsports-2016-096581 66  Bahr R, Clarsen B, Ekstrand J. Why we should focus on the burden of injuries and illnesses, not just their incidence. Br J Sports Med 2018;52:1018-1021. doi:10.1136/bjsports-2017-098160 67  Fuller CW. A kinetic model describing injury-burden in team sports. Sports Med 2017;47:2641-51. doi:10.1007/s40279-017-0746-7 68  Fletcher R, Fletcher S, Wagner E. Clinical epidemiology: the essentials. Baltimore: Wilkins & Wilkins 1982.  69  Meeuwisse WH. Assessing causation in sport injury: a multifactorial model. Clin J Sport Med 1994;4:166–170. 70  Meeuwisse WH. Athletic injury etiology: distinguishing between interaction and confounding. J Sport Med 1994;4:171–5. 71  Bahr R, Holme I. Risk factors for sports injuries — a methodological approach. Br J Sports Med 2003;37:384–92. doi:10.1136/bjsm.37.5.384 72  Hewett TE, Myer GD, Ford KR, et al. Biomechanical measures of neuromuscular control and valgus loading of the knee predict anterior cruciate ligament injury risk in female athletes a prospective study. Am J Sports Med 2005;33:492–501. doi:10.1177/0363546504269591 73  Myer GD, Ford KR, McLean SG, et al. The effects of plyometric versus dynamic stabilization and balance training on lower extremity biomechanics. Am J Sports Med 2006;34:445–55. doi:10.1177/0363546505281241 74  Drew MK, Finch CF. The relationship between training load and injury, illness and soreness: a systematic and literature review. Sports Med 2016;46:861-83. doi:10.1007/s40279-015-0459-8 75  Gabbett TJ, Whyte DG, Hartwig TB, et al. The relationship between workloads, physical performance, injury and illness in adolescent male football players. Sports Med 2014;44:989–1003. doi:10.1007/s40279-014-0179-5 76  Jones CM, Griffiths PC, Mellalieu SD. Training load and fatigue marker associations with injury and illness: a systematic review of longitudinal studies. Sports Med 2016;47:943-74 doi:10.1007/s40279-016-0619-5 164  77  Bourdon PC, Cardinale M, Murray A, et al. Monitoring athlete training loads: consensus statement. Int J Sports Physiol Perform 2017;12:S2-161. doi:10.1123/IJSPP.2017-0208 78  Smith DJ. A framework for understanding the training process leading to elite performance. Sports Med 2003;33:1103–26. 79  Foster C, Rodriguez-Marroyo JA, de Koning JJ. Monitoring training loads: the past, the present, and the future. Int J Sports Physiol Perform 2017;S2:2-8. doi:10.1123/ijspp.2016-0388 80  Szabo S, Tache Y, Somogyi A. The legacy of Hans Selye and the origins of stress research: a retrospective 75 years after his landmark brief “letter” to the editor of nature. Stress 2012;15:472–8. doi:10.3109/10253890.2012.710919 81  Selye H. A syndrome produced by diverse nocuous agents. Nature 1936;138:3479. 82  Kiely J. Periodization paradigms in the 21st century: evidence-led or tradition-driven? Int J Sports Physiol Perform 2012;7:242–50. 83  Kiely J. Periodization Theory: Confronting an inconvenient truth. Sports Med 2018;48:753-764. doi:10.1007/s40279-017-0823-y 84  Cunanan AJ, DeWeese BH, Wagle JP, et al. The general adaptation syndrome: a foundation for the concept of periodization. Sports Med 2018;48:787-797 doi:10.1007/s40279-017-0855-3 85  McArdle WD, Katch FI, Katch VL. Exercise Physiology: Nutrition, Energy, and Human Performance (Point. Seventh, North American Edition. Lippincott Williams & Wilkins 2009.  86  Bompa T, Haff GG. Periodization-5th Edition: Theory and Methodology of Training. 5th ed. Champaign, IL: Human Kinetics 2009.  87  Carfagno DG, Hendrix JC. Overtraining syndrome in the athlete: current clinical practice. Curr Sports Med Rep 2014;13:45–51. doi:10.1249/JSR.0000000000000027 88  Meeusen R, Duclos M, Foster C, et al. Prevention, diagnosis, and treatment of the overtraining syndrome: joint consensus statement of the European College of Sport Science and the American College of Sports Medicine. Med Sci Sports Exerc 2013;45:186–205. doi:10.1249/MSS.0b013e318279a10a 89  Schwellnus M, Soligard T, Alonso J-M, et al. How much is too much? (Part 2) International Olympic Committee consensus statement on load in sport and risk of illness. Br J Sports Med 2016;50:1043–52. doi:10.1136/bjsports-2016-096572 90  Coffey VG, Hawley DJA. The molecular bases of training adaptation. Sports Med 2007;37:737–63. doi:10.2165/00007256-200737090-00001 165  91  Coffey VG, Hawley JA. Training for performance: insights from molecular biology. Int J Sports Physiol Perform 2006;1:284–92. doi:10.1123/ijspp.1.3.284 92  Hawley JA, Hargreaves M, Joyner MJ, et al. Integrative biology of exercise. Cell 2014;159:738–49. doi:10.1016/j.cell.2014.10.029 93  Egan B, Zierath JR. Exercise metabolism and the molecular regulation of skeletal muscle adaptation. Cell Metab 2013;17:162–84. doi:10.1016/j.cmet.2012.12.012 94  Mann TN, Lamberts RP, Lambert MI. High responders and low responders: factors associated with individual variation in response to standardized training. Sports Med 2014;44:1113–24. doi:10.1007/s40279-014-0197-3 95  Baum A, Contrada R. The handbook of stress science: Biology, psychology, and health. Springer Publishing Company 2010.  96  Halson SL. Monitoring training load to understand fatigue in athletes. Sports Med 2014;44:139–47. doi:10.1007/s40279-014-0253-z 97  Impellizzeri FM, Rampinini E, Marcora SM. Physiological assessment of aerobic training in soccer. J Sports Sci 2005;23:583–92. doi:10.1080/02640410400021278 98  Akenhead R, Nassis GP. Training load and player monitoring in high-level football: current practice and perceptions. Int J Sports Physiol Perform 2016;11:587-93. doi:10.1123/ijspp.2015-0331 99  Chambers R, Gabbett TJ, Cole MH, et al. The use of wearable microsensors to quantify sport-specific movements. Sports Med 2015;45:1065–81. doi:10.1007/s40279-015-0332-9 100  Gabbett TJ, Nassis GP, Oetter E, et al. The athlete monitoring cycle: a practical guide to interpreting and applying training monitoring data. Br J Sports Med 2017;51:1451-52. doi:10.1136/bjsports-2016-097298 101  Borresen J, Lambert PMI. The quantification of training load, the training response and the effect on performance. Sports Med 2012;39:779–95. doi:10.2165/11317780-000000000-00000 102  Brink MS, Frencken W GP, Jordet G, et al. Coaches’ and players’ perceptions of training dose: not a perfect match. Int J Sports Physiol Perform 2014;9:497–502. doi:10.1123/IJSPP.2013-0009 103  Mujika I, Padilla S. Scientific bases for precompetition tapering strategies. Med Sci Sports Exerc 2003;35:1182–7. doi:10.1249/01.MSS.0000074448.73931.11 104  Le Meur Y, Hausswirth C, Mujika I. Tapering for competition: A review. Sci Sports 2012;27:77–87. doi:10.1016/j.scispo.2011.06.013 166  105  Banister E, Calvert T, Savage M, et al. A systems model of training for athletic performance. Aust J Sports Med 1975;7:57–61. 106  Busso T, Häkkinen K, Pakarinen A, et al. A systems model of training responses and its relationship to hormonal responses in elite weight-lifters. Eur J Appl Physiol 1990;61:48–54. 107  Wallace LK, Slattery KM, Coutts AJ. A comparison of methods for quantifying training load: relationships between modelled and actual training responses. Eur J Appl Physiol 2013;114:11–20. doi:10.1007/s00421-013-2745-1 108  Hellard P, Avalos M, Lacoste L, et al. Assessing the limitations of the Banister model in monitoring training. J Sports Sci 2006;24:509–20. doi:10.1080/02640410500244697 109  Halson SL, Jeukendrup AE. Does overtraining exist? An analysis of overreaching and overtraining research. Sports Med 2004;34:967–81. 110  Lamberts RP, Rietjens GJ, Tijdink HH, et al. Measuring submaximal performance parameters to monitor fatigue and predict cycling performance: a case study of a world-class cyclo-cross cyclist. Eur J Appl Physiol 2010;108:183–90. 111  Saw AE, Main LC, Gastin PB. Monitoring the athlete training response: subjective self-reported measures trump commonly used objective measures: a systematic review. Br J Sports Med 2016;50:281–91. doi:10.1136/bjsports-2015-094758 112  Gabbett TJ. The training-injury prevention paradox: should athletes be training smarter and harder? Br J Sports Med 2016;50:273-280. doi:10.1136/bjsports-2015-095788 113  Hulin BT, Gabbett TJ, Lawson DW, et al. The acute:chronic workload ratio predicts injury: high chronic workload may decrease injury risk in elite rugby league players. Br J Sports Med 2016;50:231-236.doi:10.1136/bjsports-2015-094817 114  Malone S, Roe M, Doran DA, et al. High chronic training loads and exposure to bouts of maximal velocity running reduce injury risk in elite Gaelic football. J Sci Med Sport Published Online First: 2016;20:250-54. doi:10.1016/j.jsams.2016.08.005 115  Gabbett TJ. The development and application of an injury prediction model for noncontact, soft-tissue injuries in elite collision sport athletes. J Strength Cond Res 2010;24:2593–603. doi:10.1519/JSC.0b013e3181f19da4 116  Impellizzeri FM, Marcora SM. Test validation in sport physiology: lessons learned from clinimetrics. Int J Sports Physiol Perform 2009;4:269–77. doi:10.1123/ijspp.4.2.269 117  Krosshaug T, Verhagen E. Investigating injury risk factors and mechanisms. In: Verhagen E, van Mechelen W, eds. Sports Injury Research. New York: Oxford University Press 2010. 109–24. 167  118  Davidson M, Keating J. Patient-reported outcome measures (PROMs): how should I interpret reports of measurement properties? A practical guide for clinicians and researchers who are not biostatisticians. Br J Sports Med 2014;48:792–6. doi:10.1136/bjsports-2012-091704 119  Thorpe RT, Atkinson G, Drust B, et al. Monitoring fatigue status in elite team-sport athletes: implications for practice. Int J Sports Physiol Perform 2017;12:S2-27. doi:10.1123/ijspp.2016-0434 120  Thorpe RT, Strudwick AJ, Buchheit M, et al. The tracking of morning fatigue status across in-season training weeks in elite soccer players. Int J Sports Physiol Perform 2016;11:947-52. doi:10.1123/ijspp.2015-0490 121  Messick S. Foundations of Validity: Meaning and consequences in psychological assessment. ETS Res Rep Ser 1993;2:i–18. doi:10.1002/j.2333-8504.1993.tb01562.x 122  Messick S. Validity of test interpretation and use. ETS Res Rep Ser 1990;1:1487–95. doi:10.1002/j.2333-8504.1990.tb01343.x 123  Messick S. The standard problem: Meaning and values in measurement and evaluation. Am Psychol 1975;30:955–66. doi:10.1037/0003-066X.30.10.955 124  Hubley AM, Zumbo BD. Validity and the consequences of test interpretation and use. Soc Indic Res 2011;103:219–30. 125  Kane MT. An argument-based approach to validation. Psychol Bull 1992;112:527–35. doi: 126  Collins L. Analysis of Longitudinal Data: The integration of theoretical model, temporal design, and statistical model. Annu Rev Psychol 2006;57:505–28. doi:10.1146/annurev.psych.57.102904.190146 127  Hoffman L. Longitudinal Analysis: Modeling Within-Person Fluctuation and Change. Routledge 2014.   128  Gibbons RD, Hedeker D, DuToit S. Advances in analysis of longitudinal data. Annu Rev Clin Psychol 2010;6:79–107. doi:10.1146/annurev.clinpsy.032408.153550 129  de Haan-Rietdijk S, Kuppens P, Hamaker EL. What’s in a day? a guide to decomposing the variance in intensive longitudinal data. Front Psychol 2016;7:891 doi:10.3389/fpsyg.2016.00891 130  Walls TA, Schafer JL. Models for intensive longitudinal data. Oxford [U.K.]: Oxford University Press 2006. 168  131  Windt J, Gabbett TJ. How do training and competition workloads relate to injury? The workload—injury aetiology model. Br J Sports Med 2017;51:428-435. doi:10.1136/bjsports-2016-096040 132  Gabbett TJ. Incidence of injury in junior and senior rugby league players. Sports Med 2004;34:849–59. 133  Hrysomallis C. Injury incidence, risk factors and prevention in australian rules football. Sports Med 2013;43:339–54. doi:10.1007/s40279-013-0034-0 134  Hägglund M, Waldén M, Magnusson H, et al. Injuries affect team performance negatively in professional football: an 11-year follow-up of the UEFA Champions League injury study. Br J Sports Med 2013;47:738–42. doi:10.1136/bjsports-2013-092215 135  Saxon L, Finch DC, Bass S. Sports participation, sports injuries and osteoarthritis. Sports Med 2012;28:123–35. doi:10.2165/00007256-199928020-00005 136  Hulin BT, Gabbett TJ, Blanch P, et al. Spikes in acute workload are associated with increased injury risk in elite cricket fast bowlers. Br J Sports Med 2014;48:708–12. doi:10.1136/bjsports-2013-092524 137  Andrew N, Gabbe BJ, Cook J, et al. Could targeted exercise programmes prevent lower limb injury in community australian football? Sports Med 2013;43:751–63. doi:10.1007/s40279-013-0056-7 138  Colby MJ, Dawson B, Heasman J, et al. Accelerometer and GPS-derived running loads and injury risk in elite australian footballers: J Strength Cond Res 2014;28:2244–52. doi:10.1519/JSC.0000000000000362 139  Veugelers KR, Young WB, Fahrner B, et al. Different methods of training load quantification and their relationship to injury and illness in elite Australian football. J Sci Med Sport 2015;19:24–8. doi:10.1016/j.jsams.2015.01.001 140  Rogalski B, Dawson B, Heasman J, et al. Training and game loads and injury risk in elite Australian footballers. J Sci Med Sport 2013;16:499–503. doi:10.1016/j.jsams.2012.12.004 141  Anderson L, Triplett-McBride T, Foster C, et al. Impact of training patterns on incidence of illness and injury during a women’s collegiate basketball season. J Strength Cond Res 2003;17:734–8. 142  Gabbett TJ. Influence of training and match intensity on injuries in rugby league. J Sports Sci 2004;22:409–17. doi:10.1080/02640410310001641638 143  Gabbett TJ, Jenkins DG. Relationship between training load and injury in professional rugby league players. J Sci Med Sport 2011;14:204–9. doi:10.1016/j.jsams.2010.12.002 169  144  Orchard JW, James T, Portus M, et al. Fast bowlers in cricket demonstrate up to 3- to 4-week delay between high workloads and increased risk of injury. Am J Sports Med 2009;37:1186–92. doi:10.1177/0363546509332430 145  Dennis R, Farhart R, Goumas C, et al. Bowling workload and the risk of injury in elite cricket fast bowlers. J Sci Med Sport 2003;6:359–67. doi:10.1016/S1440-2440(03)80031-2 146  McIntosh AS. Risk compensation, motivation, injuries, and biomechanics in competitive sport. Br J Sports Med 2005;39:2–3. doi:10.1136/bjsm.2004.016188 147  Verhagen E, Stralen MM van, Mechelen W van. Behaviour, the key factor for sports injury prevention. Sports Med 2012;40:899–906. doi:10.2165/11536890-000000000-00000 148  Bosquet L, Montpetit J, Arvisais D, et al. Effects of tapering on performance: a meta-analysis. Med Sci Sports Exerc 2007;39:1358–65. doi:10.1249/mss.0b013e31806010e0 149  Busso T, Carasso C, Lacour JR. Adequacy of a systems structure in the modeling of training effects on performance. J Appl Physiol 1991;71:2044–9. 150  Bouchard C. Genomic predictors of trainability. Exp Physiol 2012;97:347–52. doi:10.1113/expphysiol.2011.058735 151  Bouchard C, Sarzynski MA, Rice TK, et al. Genomic predictors of the maximal O2 uptake response to standardized exercise training programs. J Appl Physiol 2011;110:1160–70. doi:10.1152/japplphysiol.00973.2010 152  Gamble P. Periodization of training for team sports athletes. Strength Cond J 2006;28:56–66. 153  Flatt AA, Esco MR. Smartphone-derived heart-rate variability and training load in a women’s soccer team. Int J Sports Physiol Perform 2015;10:994–1000. doi:10.1123/ijspp.2014-0556 154  Borresen J, Lambert MI. Changes in heart rate recovery in response to acute changes in training load. Eur J Appl Physiol 2007;101:503–11. doi:10.1007/s00421-007-0516-6 155  Laux P, Krumm B, Diers M, et al. Recovery–stress balance and injury risk in professional football players: a prospective study. J Sports Sci 2015;33:2140–8. doi:10.1080/02640414.2015.1064538 156  Gabbett HT, Windt J, Gabbett TJ. Cost-benefit analysis underlies training decisions in elite sport. Br J Sports Med 2016;50:1291-1292. doi:10.1136/bjsports-2016-096079 157  Killen NM, Gabbett TJ, Jenkins DG. Training loads and incidence of injury during the preseason in professional rugby league players: J Strength Cond Res 2010;24:2079–84. doi:10.1519/JSC.0b013e3181ddafff 170  158  Gabbett TJ. Reductions in pre-season training loads reduce training injury rates in rugby league players. Br J Sports Med 2004;38:743–9. doi:10.1136/bjsm.2003.008391 159  Gabbett TJ, Domrow N. Relationships between training load, injury, and fitness in sub-elite collision sport athletes. J Sports Sci 2007;25:1507–19. doi:10.1080/02640410701215066 160  Gabbett TJ, Ullah S. Relationship between running loads and soft-tissue injury in elite team sport athletes. J Strength Cond Res 2012;26:953–60. doi:10.1519/JSC.0b013e3182302023 161  Lyman S, Fleisig GS, Waterbor JW, et al. Longitudinal study of elbow and shoulder pain in youth baseball pitchers. Med Sci Sports Exerc 2001;33:1803–10. 162  Olsen SJ, Fleisig GS, Dun S, et al. Risk factors for shoulder and elbow injuries in adolescent baseball pitchers. Am J Sports Med 2006;34:905–12. doi:10.1177/0363546505284188 163  Bartolomei S, Hoffman JR, Merni F, et al. A comparison of traditional and block periodized strength training programs in trained athletes. J Strength Cond Res 2014;28:990–7. doi:10.1519/JSC.0000000000000366 164  Tønnessen E, Shalfawi SAI, Haugen T, et al. The effect of 40-m repeated sprint training on maximum sprinting speed, repeated sprint speed endurance, vertical jump, and aerobic capacity in young elite male soccer players. J Strength Cond Res 2011;25:2364–70. doi:10.1519/JSC.0b013e3182023a65 165  Gabbett TJ. Changes in physiological and anthropometric characteristics of rugby league players during a competitive season. J Strength Cond Res 2005;19:400–8. 166  Grant JA, Bedi A, Kurz J, et al. Ability of preseason body composition and physical fitness to predict the risk of injury in male collegiate hockey players. Sports Health 2015;7:45–51. doi:10.1177/1941738114540445 167  Gabbett TJ, Ullah S, Finch CF. Identifying risk factors for contact injury in professional rugby league players--application of a frailty model for recurrent injury. J Sci Med Sport 2012;15:496–504. doi:10.1016/j.jsams.2012.03.017 168  Hulin B, Gabbett T, Caputi P, et al. Low chronic workload and the acute:chronic workload ratio are more predictive of injury than between-match recovery time: A two-season prospective cohort study in elite rugby league players. Br J Sports Med 2016;50:1008-1012.  doi:10.1136/bjsports-2015-095364 169  Ehrmann FE, Duncan CS, Sindhusake D, et al. GPS and injury prevention in professional soccer. J Strength Cond Res 2016;30:360–367. doi:10.1519/JSC.0000000000001093 170  Blanch P, Gabbett TJ. Has the athlete trained enough to return to play safely? The acute:chronic workload ratio permits clinicians to quantify a player’s risk of subsequent injury. Br J Sports Med 2016;50:471-475. doi:10.1136/bjsports-2015-095445 171  171  Emery CA. Injury prevention in paediatric sport-related injuries: a scientific approach. Br J Sports Med 2010;44:64–9. doi:10.1136/bjsm.2009.068353 172  Gastin PB, Meyer D, Huntsman E, et al. Increase in injury risk with low body mass and aerobic-running fitness in elite Australian football. Int J Sports Physiol Perform 2015;10:458–63. doi:10.1123/ijspp.2014-0257 173  Windt J, Gabbett TJ, Ferris D, et al. Training load--injury paradox: is greater preseason participation associated with lower in-season injury risk in elite rugby league players? Br J Sports Med 2017;51:645-650. doi:10.1136/bjsports-2016-095973 174  Drew MK, Purdam C. Time to bin the term ‘overuse’ injury: is ‘training load error’ a more accurate term? Br J Sports Med 2016;50:1423-1424. doi:10.1136/bjsports-2015-095543 175  Timpka T, Jacobsson J, Dahlström Ö, et al. The psychological factor ‘self-blame’ predicts overuse injury among top-level Swedish track and field athletes: a 12-month cohort study. Br J Sports Med 2015;49:1472–7. doi:10.1136/bjsports-2015-094622 176  Orchard JW, Blanch P, Paoloni J, et al. Cricket fast bowling workload patterns as risk factors for tendon, muscle, bone and joint injuries. Br J Sports Med 2015;49:1064–8. doi:10.1136/bjsports-2014-093683 177  Raudenbush SW, Bryk AS. Hierarchical linear models: applications and data analysis methods. 2nd ed. Thousand Oaks, CA: Sage Publications 2002.  178  Hox JJ. Multilevel analysis of grouped and longitudinal data. Psychology Press. 2000;15–32. 179  Windt J, Zumbo BD, Sporer B, et al. Why do workload spikes cause injuries, and which athletes are at higher risk? Mediators and moderators in workload–injury investigations. Br J Sports Med 2017;51:993-994. doi:10.1136/bjsports-2016-097255 180  Bittencourt NFN, Meeuwisse WH, Mendonça LD, et al. Complex systems approach for sports injuries: moving from risk factor identification to injury pattern recognition—narrative review and new concept. Br J Sports Med 2016;50:1309-14. doi:10.1136/bjsports-2015-095850 181  Wu AD, Zumbo BD. Understanding and using mediators and moderators. Soc Indic Res 2007;87:367. doi:10.1007/s11205-007-9143-1 182  Malone S, Roe M, Doran DA, et al. Aerobic fitness and playing experience protect against spikes in workload: the role of the acute:chronic workload ratio on injury risk in elite Gaelic football. Int J Sports Physiol Perform 2016;24:1–25. doi:10.1123/ijspp.2016-0090 183  Bahr R. Why screening tests to predict injury do not work—and probably never will…: a critical review. Br J Sports Med 2016;50:776–80. doi:10.1136/bjsports-2016-096256 172  184  Windt J, Ardern CL, Gabbett TJ, et al. Getting the most out of intensive longitudinal data: A methodological review of workload—injury studies. BMJ Open 2018; In Press. 185  Ginexi EM, Riley W, Atienza AA, et al. The promise of intensive longitudinal data capture for behavioral health research. Nicotine Tob Res 2014;16:S73–5. doi:10.1093/ntr/ntt273 186  Hamaker EL, Wichers M. No time like the present: discovering the hidden dynamics in intensive longitudinal data. Curr Dir Psychol Sci 2017;26:10–5. doi:10.1177/0963721416666518 187  McCall A, Carling C, Nedelec M, et al. Risk factors, testing and preventative strategies for non-contact injuries in professional football: current perceptions and practices of 44 teams from various premier leagues. Br J Sports Med 2014;48:1352–7. doi:10.1136/bjsports-2014-093439 188  Black GM, Gabbett TJ, Cole MH, et al. Monitoring workload in throwing-dominant sports: a systematic review. Sports Med 2016;10:1–14. doi:10.1007/s40279-016-0529-6 189  Keselman HJ, Huberty CJ, Lix LM, et al. Statistical practices of educational researchers: an analysis of their ANOVA, MANOVA, and ANCOVA analyses. Rev Educ Res 1998;68:350–86. doi:10.2307/1170601 190  Weir A, Rabia S, Ardern C. Trusting systematic reviews and meta-analyses: all that glitters is not gold! Br J Sports Med 2016;50:1100-01. doi:10.1136/bjsports-2015-095896 191  Lemon SC, Wang ML, Haughton CF, et al. Methodological quality of behavioural weight loss studies: a systematic review. Obes Rev 2016;7:636-44. doi:10.1111/obr.12412 192  Olivier B, Taljaard T, Burger E, et al. Which extrinsic and intrinsic factors are associated with non-contact injuries in adult cricket fast bowlers? Sports Med 2016;46:79-101. doi:10.1007/s40279-015-0383-y 193  Whittaker JL, Small C, Maffey L, et al. Risk factors for groin injury in sport: an updated systematic review. Br J Sports Med 2015;49:803–9. doi:10.1136/bjsports-2014-094287 194  Hulme A, Salmon PM, Nielsen RO, et al. Closing Pandora’s Box: adapting a systems ergonomics methodology for better understanding the ecological complexity underpinning the development and prevention of running-related injury. Theor Issues Ergon Sci 2017;18:338–59. doi:10.1080/1463922X.2016.1274455 195  Cook C. Predicting future physical injury in sports: it’s a complicated dynamic system. Br J Sports Med 2016;50:1356-57. doi:10.1136/bjsports-2016-096445 196  Sun L, Wu R. Mapping complex traits as a dynamic system. Phys Life Rev 2015;13:155–85. doi:10.1016/j.plrev.2015.02.007 173  197  Wilkinson M, Akenhead R. Violation of statistical assumptions in a recent publication? Int J Sports Med 2013;34:281. doi:10.1055/s-0032-1331775 198  Brooks JHM, Fuller CW, Kemp SPT, et al. An assessment of training volume in professional rugby union and its impact on the incidence, severity, and nature of match and training injuries. J Sports Sci 2008;26:863–73. doi:10.1080/02640410701832209 199  Duhig S, Shield AJ, Opar D, et al. Effect of high-speed running on hamstring strain injury risk. Br J Sports Med 2016;50:1536-40. doi:10.1136/bjsports-2015-095679 200  Murray NB, Gabbett TJ, Townshend AD, et al. Individual and combined effects of acute and chronic running loads on injury risk in elite Australian footballers. Scand J Med Sci Sports 2017;27:990–8. doi:10.1111/sms.12719 201  Murray NB, Gabbett TJ, Townshend AD. Relationship between pre-season training load and in-season availability in elite australian football players. Int J Sports Physiol Perform 2017;12;749-55. doi:10.1123/ijspp.2015-0806 202  Dennis R, Farhart P, Clements M, et al. The relationship between fast bowling workload and injury in first-class cricketers: a pilot study. J Sci Med Sport Sports Med Aust 2004;7:232–6. 203  Dennis RJ, Finch CF, Farhart PJ. Is bowling workload a risk factor for injury to Australian junior cricket fast bowlers? Br J Sports Med 2005;39:843–6. doi:10.1136/bjsm.2005.018515 204  Saw R, Dennis RJ, Bentley D, et al. Throwing workload and injury risk in elite cricketers. Br J Sports Med 2011;45:805–8. doi:10.1136/bjsm.2009.061309 205  Bresciani G, Cuevas MJ, Garatachea N, et al. Monitoring biological and psychological measures throughout an entire season in male handball players. Eur J Sport Sci 2010;10:377–84. doi:10.1080/17461391003699070 206  Arnason A, Sigurdsson SB, Gudmundsson A, et al. Risk factors for injuries in football. Am J Sports Med 2004;32:5S-16S. doi:10.1177/0363546503258912 207  Brink MS, Nederhof E, Visscher C, et al. Monitoring load, recovery, and performance in young elite soccer players. J Strength Cond Res Natl Strength Cond Assoc 2010;24:597–603. doi:10.1519/JSC.0b013e3181c4d38b 208  Mallo J, Dellal A. Injury risk in professional football players with special reference to the playing position and training periodization. J Sports Med Phys Fitness 2012;52:631–8. 209  Clausen MB, Zebis MK, Møller M, et al. High injury incidence in adolescent female soccer. Am J Sports Med 2014;42:2487–94. doi:10.1177/0363546514541224 174  210  Owen AL, Forsyth JJ, Wong DP, et al. Heart rate-based training intensity and its impact on injury incidence among elite-level professional soccer players. J Strength Cond Res Natl Strength Cond Assoc 2015;29:1705–12. doi:10.1519/JSC.0000000000000810 211  Bowen L, Gross AS, Gimpel M, et al. Accumulated workloads and the acute:chronic workload ratio relate to injury risk in elite youth football players. Br J Sports Med 2016;51:452-9. doi:10.1136/bjsports-2015-095820 212  Malisoux L, Frisch A, Urhausen A, et al. Monitoring of sport participation and injury risk in young athletes. J Sci Med Sport 2013;16:504–8. doi:10.1016/j.jsams.2013.01.008 213  Visnes H, Bahr R. Training volume and body composition as risk factors for developing jumper’s knee among young elite volleyball players. Scand J Med Sci Sports 2013;23:607–13. doi:10.1111/j.1600-0838.2011.01430.x 214  Cohen J. Statistical power analysis for the behavioral sciences. 2nd. Hillsdale, NJ: erlbaum 1988.  215  Kraemer HC, Thiemann S. How many subjects. Sage 1987.  216  Kutner MH, Nachtsheim C, Neter J. Applied linear regression models. McGraw-Hill/Irwin 2004.  217  Silvey SD. Multicollinearity and imprecise estimation. J Royal Stat Soc 1969;31:539–52. 218  Zidek JV, Wong H, Le ND, et al. Causality, measurement error and multicollinearity in epidemiology. Environmetrics 1996;7:441–451. 219  Datson N, Hulton A, Andersson H, et al. Applied physiology of female soccer: an update. Sports Med 2014;44:1225–40. doi:10.1007/s40279-014-0199-1 220  Nielsen RO, Bertelsen ML, Møller M, et al. Training load and structure-specific load: applications for sport injury causality and data analyses. Br J Sports Med 2018;52:1016-1017. doi:10.1136/bjsports-2017-097838 221  Corraini P, Olsen M, Pedersen L, et al. Effect modification, interaction and mediation: an overview of theoretical insights for clinical investigators. Clin Epidemiol 2017;9:331–8. doi:10.2147/CLEP.S129728 222  VanderWeele TJ. On the distinction between interaction and effect modification: Epidemiology 2009;20:863–71. doi:10.1097/EDE.0b013e3181ba333c 223  Ghisletta P, Spini D. An introduction to generalized estimating equations and an application to assess selectivity effects in a longitudinal study on very old individuals. J Educ Behav Stat 2004;29:421–37. doi:10.3102/10769986029004421 175  224  Bellera CA, MacGrogan G, Debled M, et al. Variables with time-varying effects and the Cox model: Some statistical concepts illustrated with a prognostic factor study in breast cancer. BMC Med Res Methodol 2010;10:20. doi:10.1186/1471-2288-10-20 225  Hulme A, Finch CF. From monocausality to systems thinking: a complementary and alternative conceptual approach for better understanding the development and prevention of sports injury. Inj Epidemiol 2015;2:31. doi:10.1186/s40621-015-0064-1 226  Ward P, Coutts AJ, Pruna R, et al. Putting the ‘i’ back in team. Int J Sports Physiol Perform 2018. 1–14. doi:10.1123/ijspp.2018-0154 227  Buchheit M. Applying the acute:chronic workload ratio in elite football: worth the effort? Br J Sports Med 2016;51:1325-27. doi:10.1136/bjsports-2016-097017 228  Newgard CD, Lewis RJ. Missing data: how to best account for what is not known. JAMA 2015;314:940–1. doi:10.1001/jama.2015.10516 229  El-Masri MM, Fox-Wasylyshyn SM. Missing data: an introductory conceptual overview for the novice researcher. Can J Nurs Res 2005;37:156–71. 230  Hippel PTV. How to impute interactions, squares, and other transformed variables. Sociol Methodol 2009;39:265–91. doi:10.1111/j.1467-9531.2009.01215.x 231  McDonald JH. Handbook of biological statistics. Sparky House Publishing Baltimore, MD 2009. 232  Therneau TM, Grambsch PM. Modeling survival data: extending the Cox model. New York: Springer 2000.  233  Hill AB. The Environment and Disease: Association or Causation? Proc R Soc Med 1965;58:295–300. 234  Höfler M. The Bradford Hill considerations on causality: a counterfactual perspective. Emerg Themes Epidemiol 2005;2:11. doi:10.1186/1742-7622-2-11 235  Rothman KJ, Greenland S. Causation and causal inference in epidemiology. Am J Public Health 2005;95 Suppl 1:S144-150. doi:10.2105/AJPH.2004.059204 236  Carey DL, Blanch P, Ong K-L, et al. Training loads and injury risk in Australian football—differing acute: chronic workload ratios influence match injury risk. Br J Sports Med 2017;51:1215-1220. doi:10.1136/bjsports-2016-096309 237  Tan X, Shiyko MP, Li R, et al. A time-varying effect model for intensive longitudinal data. Psychol Methods 2012;17:61–77. doi:10.1037/a0025814 176  238  Hulme A, Thompson J, Nielsen RO, et al. Towards a complex systems approach in sports injury research: simulating running-related injury development with agent-based modelling. Br J Sports Med 2018. doi:10.1136/bjsports-2017-098871 239  Carey DL, Ong K-L, Whiteley R, et al. Predictive modelling of training loads and injury in Australian football. Int J Comp Sci Sp 2018;1:49-66. 240  Colby MJ, Dawson B, Peeling P, et al. Multivariate modelling of subjective and objective monitoring data improve the detection of non-contact injury risk in elite Australian footballers. J Sci Med Sport 2017;20:1068-1074. 10.1016/j.jsams.2017.05.010 241  Møller M, Nielsen RO, Attermann J, et al. Handball load and shoulder injury rate: a 31-week cohort study of 679 elite youth handball players. Br J Sports Med 2017;51:231–7. doi:10.1136/bjsports-2016-096927 242  Rossi A, Pappalardo L, Cintia P, et al. Effective injury prediction in professional soccer with GPS data and machine learning. ArXiv170508079 Stat Published Online First: 23 May 2017. 243  Stares J, Dawson B, Peeling P, et al. Identifying high risk loading conditions for in-season injury in elite Australian football players. J Sci Med Sport 2018;21:46-51. doi:10.1016/j.jsams.2017.05.012 244  Williams S, Trewartha G, Cross MJ, et al. Monitoring what matters: a systematic process for selecting training load measures. Int J Sports Physiol Perform 2017;12:S2-101. doi:10.1123/ijspp.2016-0337 245  Williams S, Booton T, Watson M, et al. Heart rate variability is a moderating factor in the workload-injury relationship of competitive Crossfittm athletes. J Sports Sci Med 2017;16:443–9. 246  Sampson JA, Murray A, Williams S, et al. Injury risk-workload associations in NCAA American college football. J Sci Med Sport Published Online First: 22 May 2018. doi:10.1016/j.jsams.2018.05.019 247  Garcia TP, Marder K. Statistical approaches to longitudinal data analysis in neurodegenerative diseases: huntington’s disease as a model. Curr Neurol Neurosci Rep 2017;17:14. doi:10.1007/s11910-017-0723-4 248  Yang J, Zaitlen NA, Goddard ME, et al. Advantages and pitfalls in the application of mixed-model association methods. Nat Genet 2014;46:100–6. doi:10.1038/ng.2876 249  Russell MA, Almeida DM, Maggs JL. Stressor-related drinking and future alcohol problems among university students. Psychol Addict Behav 2017;31:676–87. doi:10.1037/adb0000303 177  250  Bolger N, Laurenceau J-P. Intensive Longitudinal Methods: An Introduction to Diary and Experience Sampling Research. Guilford Press 2013.  251  Gelman A, Hill J. Data analysis using regression and multilevel/hierarchical models. New York: Cambridge University Press 2007. (accessed 12 Apr 2016). 252  Horton NJ, Switzer SS. Statistical methods in the journal. N Engl J Med 2005;353:1977–9. doi:10.1056/NEJM200511033531823 253  Nielsen RØ, Malisoux L, Møller M, et al. Shedding light on the etiology of sports injuries: a look behind the scenes of time-to-event analyses. J Orthop Sports Phys Ther 2016;46:300–11. doi:10.2519/jospt.2016.6510 254  Wu L, Liu W, Yi GY, et al. Analysis of longitudinal and survival data: joint modeling, inference methods, and issues. J Prob Stat. 2012;2012:1-17. doi:10.1155/2012/640153 255  Andersen PK, Syriopoulou E, Parner ET. Causal inference in survival analysis using pseudo-observations. Stat Med 2017;36:2669–81. doi:10.1002/sim.7297 256  Hansen SN, Andersen PK, Parner ET. Events per variable for risk differences and relative risks using pseudo-observations. Lifetime Data Anal N Y 2014;20:584–98. doi:10.1007/s10985-013-9290-4 257  Hickey GL, Philipson P, Jorgensen A, et al. Joint modelling of time-to-event and multivariate longitudinal outcomes: recent developments and issues. BMC Med Res Methodol 2016;16. doi:10.1186/s12874-016-0212-5 258  Mahmood A, Ullah S, Finch CF. Application of survival models in sports injury prevention research: a systematic review. Br J Sports Med 2014;48:630–630. doi:10.1136/bjsports-2014-093494.190 259  Carey DL, Crossley KM, Whiteley R, et al. Modelling training loads and injuries: the dangers of discretization. Med Sci Sports Exe Published Online First: June 2018. doi:10.1249/MSS.0000000000001685 260  Maldonado G, Greenland S. The importance of critically interpreting simulation studies. Epidemiology 1997;8:453–456. 261  Trail JB, Collins LM, Rivera DE, et al. Functional data analysis for dynamical system identification of behavioral processes. Psychol Methods 2014;19:175–87. doi:10.1037/a0034035 262  Tan X, Shiyko MP, Li R, et al. A time-varying effect model for intensive longitudinal data. Psychol Methods 2012;17:61–77. doi:10.1037/a0025814   178  263  Casals M, Finch CF. Sports Biostatistician: a critical member of all sports science and medicine teams for injury prevention. Inj Prev 2017;23:423-427. doi:10.1136/injuryprev-2016-042211 264  Windt J, Ekstrand J, Khan KM, et al. Does player unavailability affect football teams’ match physical outputs? A two-season study of the UEFA champions league. J Sci Med Sport 2018;21:525-532. doi:10.1016/j.jsams.2017.08.007 265  Ekstrand J, Waldén M, Hägglund M. Hamstring injuries have increased by 4% annually in men’s professional football, since 2001: a 13-year longitudinal analysis of the UEFA Elite Club injury study. Br J Sports Med 2016;50:731–7. doi:10.1136/bjsports-2015-095359 266  Ekstrand J, Hägglund M, Kristenson K, et al. Fewer ligament injuries but no preventive effect on muscle injuries and severe injuries: an 11-year follow-up of the UEFA Champions League injury study. Br J Sports Med 2013;47:732–7. doi:10.1136/bjsports-2013-092394 267  Rampinini E, Impellizzeri FM, Castagna C, et al. Technical performance during soccer matches of the Italian Serie A league: Effect of fatigue and competitive level. J Sci Med Sport 2009;12:227–33. doi:10.1016/j.jsams.2007.10.002 268  Dupont G, Nedelec M, McCall A, et al. Effect of 2 soccer matches in a week on physical performance and injury rate. Am J Sports Med 2010;38:1752–8. doi:10.1177/0363546510361236 269  Gregson W, Drust B, Atkinson G, et al. Match-to-match variability of high-speed activities in premier league soccer. Int J Sports Med 2010;31:237–42. doi:10.1055/s-0030-1247546 270  Di Salvo V, Baron R, Tschan H, et al. Performance characteristics according to playing position in elite soccer. Int J Sports Med 2007;28:222–7. doi:10.1055/s-2006-924294 271  Bradley PS, Carling C, Archer D, et al. The effect of playing formation on high-intensity running and technical profiles in English FA Premier League soccer matches. J Sports Sci 2011;29:821–30. doi:10.1080/02640414.2011.561868 272  Di Salvo V, Gregson W, Atkinson G, et al. Analysis of high intensity activity in Premier League soccer. Int J Sports Med 2009;30:205–12. doi:10.1055/s-0028-1105950 273  Bradley PS, Noakes TD. Match running performance fluctuations in elite soccer: indicative of fatigue, pacing or situational influences? J Sports Sci 2013;31:1627–38. doi:10.1080/02640414.2013.796062 274  Ekstrand J, Hägglund M, Waldén M. Injury incidence and injury patterns in professional football: the UEFA injury study. Br J Sports Med 2011;45:553–8. doi:10.1136/bjsm.2009.060582 275  Sports digital media and broadcast services firm. Deltatre. 2012. (accessed 27 Feb 2017). 179  276  Carling C. Analysis of physical activity profiles when running with the ball in a professional soccer team. J Sports Sci 2010;28:319–26. doi:10.1080/02640410903473851 277  Raudenbush SW, Bryk A., Congdon R. HLM 7.01 for Windows. Skokie, IL: Scientific Software International, Inc 2013.  278  Batterham AM, Hopkins WG. Making meaningful inferences about magnitudes. Int J Sports Physiol Perform 2006;1:50-57. (accessed 4 Apr 2016). 279  Hopkins W, Marshall S, Batterham A, et al. Progressive statistics for studies in sports medicine and exercise science. Med Sci Sports Exerc 2009;41:3. 280  Hayes A. A primer on multilevel modeling. Hum Commun Res 2006;32:385–410. doi:10.1111/j.1468-2958.2006.00281.x 281  Paul DJ, Bradley PS, Nassis GP. Factors affecting match running performance of elite soccer players: shedding some light on the complexity. Int J Sports Physiol Perform 2015;10:516–9. doi:10.1123/IJSPP.2015-0029 282  Enders CK, Tofighi D. Centering predictor variables in cross-sectional multilevel models: A new look at an old issue. Psychol Methods 2007;12:121–38. doi:10.1037/1082-989X.12.2.121 283  Agre JC, Ash N, Cameron MC, et al. Suprascapular neuropathy after intensive progressive resistive exercise: case report. Arch Phys Med Rehabil 1987;68:236–8. 284  Hopkins WG. A spreadsheet for deriving a confidence interval, mechanistic inference and clinical inference from a p value. Sportscience 2007;16–20. 285  Link D, Lorenzo MF de. Seasonal pacing - match importance affects activity in professional soccer. PLOS ONE 2016;11:e0157127. doi:10.1371/journal.pone.0157127 286  Carling C. Interpreting physical performance in professional soccer match-play: should we be more pragmatic in our approach? Sports Med 2013;43:655–63. doi:10.1007/s40279-013-0055-8 287  Torreño N, Munguía-Izquierdo D, Coutts A, et al. Relationship between external and internal load of professional soccer players during full-matches in official games using GPS and heart rate technology. Int J Sports Physiol Perform 2016;11:940-946. doi:10.1123/ijspp.2015-0252 288  Lago-Peñas C, Lago-Ballesteros J, Dellal A, et al. Game-related statistics that discriminated winning, drawing and losing teams from the Spanish Soccer League. J Sports Sci Med 2010;9:288–93. 180  289  Regenhuber M. IMPECT & Packing: the future of football analytics is here. 2016. (accessed 7 Mar 2017). 290  Lolli L, Batterham AM, Hawkins R, et al. The acute-to-chronic workload ratio: an inaccurate scaling index for an unnecessary normalisation process? Br J Sports Med 2018. doi:10.1136/bjsports-2017-098884 291  Menaspà P. Are rolling averages a good way to assess training load for injury prevention? Br J Sports Med 2017;51:618-619. doi:10.1136/bjsports-2016-096131 292  Menaspà P. Building evidence with flawed data? The importance of analysing valid data. Br J Sports Med 2017;51:1173–1173. doi:10.1136/bjsports-2016-097029 293  Lolli L, Batterham AM, Hawkins R, et al. Mathematical coupling causes spurious correlation within the conventional acute-to-chronic workload ratio calculations. Br J Sports Med 2017. doi:10.1136/bjsports-2017-098110 294  Williams S, West S, Cross MJ, et al. Better way to determine the acute:chronic workload ratio? Br J Sports Med 2017;51:209–10. doi:10.1136/bjsports-2016-096589 295  Drew MK, Blanch P, Purdam C, et al. Yes, rolling averages are a good way to assess training load for injury prevention. Is there a better way? Probably, but we have not seen the evidence. Br J Sports Med 2017;51:618–9. doi:10.1136/bjsports-2016-096609 296  Gabbett TJ, Kennelly S, Sheehan J, et al. If overuse injury is a ‘training load error’, should undertraining be viewed the same way? Br J Sports Med 2016;50:1017-1018. doi:10.1136/bjsports-2016-096308 297  Hulin BT. The never-ending search for the perfect acute:chronic workload ratio: what role injury definition? Br J Sports Med 2017;51:991-992. doi:10.1136/bjsports-2016-097279 298  Sampson JA, Fullagar HHK, Murray A. Evidence is needed to determine if there is a better way to determine the acute:chronic workload. Br J Sports Med 2017;51:621–2. doi:10.1136/bjsports-2016-097085 299  Windt J, Gabbett TJ. Is it all for naught? What does mathematical coupling mean for acute:chronic workload ratios? Br J Sports Med 2018. Online First. doi:10.1136/bjsports-2017-098925 300  Buist I, Bredeweg SW, Mechelen W van, et al. No effect of a graded training program on the number of running-related injuries in novice runners a randomized controlled trial. Am J Sports Med 2008;36:33–9. doi:10.1177/0363546507307505 301  Ahtiainen JP, Pakarinen A, Alen M, et al. Muscle hypertrophy, hormonal adaptations and strength development during strength training in strength-trained and untrained men. Eur J Appl Physiol 2003;89:555–63. 181  302  Gabbett TJ, Domrow N. Risk factors for injury in subelite rugby league players. Am J Sports Med 2005;33:428–34. doi:10.1177/0363546504268407 303  Brooks JHM, Fuller CW, Kemp SPT, et al. Epidemiology of injuries in English professional rugby union: part 2 training Injuries. Br J Sports Med 2005;39:767–75. doi:10.1136/bjsm.2005.018408 304  Delecroix B, McCall A, Dawson B, et al. Workload and non-contact injury incidence in elite football players competing in European leagues. Eur J Sport Sci 2018;0:1–8. doi:10.1080/17461391.2018.1477994 305  McCall A, Dupont G, Ekstrand J. Internal workload and non-contact injury: a one-season study of five teams from the UEFA Elite Club Injury Study. Br J Sports Med 2018. Online First. doi:10.1136/bjsports-2017-098473 306  Vanrenterghem J, Nedergaard NJ, Robinson MA, et al. Training load monitoring in team sports: a novel framework separating physiological and biomechanical load-adaptation pathways. Sports Med 2017;47:2135–42. doi:10.1007/s40279-017-0714-2 307  Bertelsen ML, Hulme A, Petersen J, et al. A framework for the etiology of running-related injuries. Scand J Med Sci Sports 2017;27:1170–1180. 308  Malone S, Hughes B, Doran DA, et al. Can the workload–injury relationship be moderated by improved strength, speed and repeated-sprint qualities? J Sci Med Sport Published Online First: 2 February 2018. doi:10.1016/j.jsams.2018.01.010 309  Malone S, Owen A, Newton M, et al. The acute:chonic workload ratio in relation to injury risk in professional soccer. J Sci Med Sport 2017;20:561–5. doi:10.1016/j.jsams.2016.10.014 310  Robertson S, Bartlett JD, Gastin PB. Red, amber, or green? Athlete monitoring in team sport: the need for decision-support systems. Int J Sports Physiol Perform 2016;12:S2-73. doi:10.1123/ijspp.2016-0541 311  Colby MJ, Dawson B, Heasman J, et al. Preseason workload volume and high-risk periods for noncontact injury across multiple australian football league seasons: J Strength Cond Res 2017;31:1821–9. doi:10.1519/JSC.0000000000001669 312  Murray A, Fullagar HHK, Delaney JA, et al. Bradford Factor and seasonal injury risk in Division I-A collegiate American footballers. Sci Med Footb 2018;1–4. doi:10.1080/24733938.2018.1452284 313  Moran LR, Hegedus EJ, Bleakley CM, et al. Jump load: capturing the next great injury analytic. Br J Sports Med 2018. doi:10.1136/bjsports-2018-099103 182  314  Skazalski C, Whiteley R, Hansen C, et al. A valid and reliable method to measure jump-specific training and competition load in elite volleyball players. Scand J Med Sci Sports 2018;28:1578–1585. 315  Lyman S, Fleisig GS, Andrews JR, et al. Effect of pitch type, pitch count, and pitching mechanics on risk of elbow and shoulder pain in youth baseball pitchers. Am J Sports Med 2002;30:463–8. 316  Bahr R. Demise of the fittest: are we destroying our biggest talents? Br J Sports Med 2014;48:1265–7. doi:10.1136/bjsports-2014-093832 317  Murray NB, Gabbett TJ, Townshend AD, et al. Calculating acute:chronic workload ratios using exponentially weighted moving averages provides a more sensitive indicator of injury likelihood than rolling averages. Br J Sports Med 2017;51:749-754. doi:10.1136/bjsports-2016-097152 318  McCall A, Fanchini M, Coutts AJ. Prediction: the modern day sports science/medicine ‘quest for the holy grail.’ Int J Sports Physiol Perform 2017. 1–11. doi:10.1123/ijspp.2017-0137 319  Rossi A, Pappalardo L, Cintia P, et al. Effective injury forecasting in soccer with GPS training data and machine learning. PLOS ONE 2018;13:e0201264. doi:10.1371/journal.pone.0201264 320  Bolling C, Mechelen W van, Pasman HR, et al. Context matters: revisiting the first step of the ‘sequence of prevention’ of sports injuries. Sports Med 2018. 1–8. doi:10.1007/s40279-018-0953-x  183  Appendices Appendix A  - Additional tables and figures from Chapter 5 Table A.1 – Example search categories and terms used in chapter 5. Category Search Terms Population Athlet*, sport*  Method Systematic review, Consensus statement Outcome (Injury) Injur*, illness*, strain, sprain, incidence, overuse, overreach*, accidents, stress, wellness, recover* Workload Training, resistance training, external load, internal load, workload, acute:chronic workload ratio, congested calendar, physical exertion, session RPE, global position systems, accelerometry, intensity, duration, physical fitness, fatigue 184   Figure A.1: Flow chart of included articles for Chapter 5.185  Table A.2: Expanded table of 3-fold alignment (with descriptions)     Themes of theoretical model  Themes of temporal design: intensive longitudinal data Statistical summary and typical uses Method n Multifactorial aetiology Between and within-athlete differences Complex system Includes time-varying and time-invariant variables Missing/unbalanced data (due to rep. measures)  Repeated measure dependency Incorporates time in the analysis Correlation (Pearson and Spearman) 10 X X X X X X X 7 of the 10 articles correlated team loads with injury incidence (team-level). Of those at the individual level, 2 looked at # of pre-season sessions and % in-season completed, and 1 looked at training load and injury subscale on the REST-Q. Anderson - 2003 Bresciani - 2010 Books - 2008 Gabbett - 2004 Gabbett Jenkins - 2011 Killen 2010 Mallo Dellal - 2012 Murray Gabbett - 2016 Owen - 2015 Windt – 2016   Can only handle one x and one y variable. Correlation could be at the team level (training load and # of injuries), or on individual level with quantitative outcome, but cannot differentiate within/between athlete differences.  No interactions between multiple predictors No, assumes independent observations. Assumes one observation per research participant/study unit, so participants couldn't have different numbers of observations. In this case the correlation assumes independent observations so dependency is not taken into account.   Could, through having time as one of the variables, but cannot account for temporality Unpaired t-test 6 X X X X X X X Generally, compared injured and uninjured players across a season. A group comparison of loading across the year fails to account for between/within athlete considerations and doesn't specify temporality. Dennis et al., 2003 Dennis et al., 2005 Duhig et al - 2016 Owen et al., 2015 Saw et al., 2011 Visnes Bahr., 2013   Can only handle one x (grouping) and one y (outcome) variable Only between-athlete differences (injured vs. uninjured) No interactions between multiple predictors Independent samples t-test compare group means on either a time-varying (e.g average workload) or time-invariant variable (e.g. height), not both. No, assumes one observation per research participant/study unit, so by design forces a balanced set (1 observation per participant) By definition, assumes independence None included time in the analysis.  186  Table A.2 (continued): Expanded table of 3-fold alignment (with descriptions)       Themes of theoretical model  Themes of temporal design: intensive longitudinal data Statistical summary and typical uses Method n Multifactorial aetiology Between and within-athlete differences Complex system Includes time-varying and time-invariant variables Missing/unbalanced data (due to rep. measures)  Repeated measure dependency Incorporates time in the analysis Chi-Square Tests 1 X X X X X X X  Murray - 2016 - IJSPP   Only examines load groups and injury incidence Examines differences in injury incidence across different load groups only No interactions between multiple predictors Included load groups and injury incidence only Forces 1 observation (aggregated variable) per participant Designed for independent observations and groups Only included load groups and injury incidence Relative Risk Calculations 8 O X X X X X X Many of these RR approaches seem to use RRs that traditionally require independence, but do not account for this in their analysis. Bowen - 2016 Dennis - 2003 Dennis - 2005 Hulin - 2014 Hulin - 2016 Hulin - 2016 Murray - 2016 - Scand.   In some cases, authors examined relative risks of loading groups after subdividing across another variable, like chronic workload, making it multifactorial. Other authors only examined risks across load groups. No differentiation, and independence assumed No interactions between multiple predictors Only loading (time-varying variables included) Assumes independence of observations Assumes independence Uses weeks as unit of analysis but no incorporation of time into the calculations 187   Table A.2 (continued): Expanded table of 3-fold alignment (with descriptions)      Themes of theoretical model  Themes of temporal design: intensive longitudinal data Statistical summary and typical uses Method n Multifactorial aetiology Between and within-athlete differences Complex system Includes time-varying and time-invariant variables Missing/unbalanced data (due to rep. measures)  Repeated measure dependency Incorporates time in the analysis Regression (logistic, linear, multinomial) 13 O X X X X X X   Arnason 2004 - Regular logistic Bowen - 2016 - Regular logistic Brink - 2010 - Regular multinomial Colby - 2014 - Regular logistic Duhig - 2016 - Regular logistic Gabbett Domrow - 2007 - Regular linear Hulin - 2014 - Regular logistic Hulin - 2016 - Regular logistic Hulin - 2016 - Regular logistic Murray - 2016 - Regular logistic (Scand) Owen - 2015 - Regular linear Rogalski - 2013 - Regular logistic Visnes Bahr - 2013 - Regular logistic   Some authors include multiple variables, others only single load measurements independently  Assume independent observations, so cannot examine within-athlete differences None included interactions between predictors  Assumes 1 observation per unit  Becomes time invariant on aggregation  In this case, some authors have included both, but in doing so violate the independence assumption  No, assumes one observation per athlete or research unit Assumes independence.   Visnes & Bahr (2013) do take it into account through logistic regression, but do not have the benefit of ILD  Assumes 1 observation per unit Paired t-test 2 X X X X X √ √   Saw et al., 2011 Dennis et al., 2003   No, only one variable (before/after) and outcome (load) Only examines within-athlete differences No interactions between multiple predictors No - only time-varying variables All subjects must have 2 'observations' By definition, accounts for repeated measures through paired sample Time is incorporated as pre-injury and 'injury' blocks of time 188  Table A.2 (continued): Expanded table of 3-fold alignment (with descriptions)          Themes of theoretical model  Themes of temporal design: intensive longitudinal data Statistical summary and typical uses Method n Multifactorial aetiology Between and within-athlete differences Complex system Includes time-varying and time-invariant variables Missing/unbalanced data (due to rep. measures)  Repeated measure dependency Incorporates time in the analysis Repeated measures ANOVA (One- or two-way ANOVA allowing for between- and within) 5 O O X O X √ √ Killen used 1-way ANOVA to compare the load in early pre-season compared to late pre-season, and used chi-square analyses to compare injury rate in the early and late pre-season. Uses the two separate analyses to tentatively link load leads to injury. Gabbett, 2004 performed a 2-way ANOVA comparing loads (season X month), so time is included.  Ehrman - 2016 Gabbett - 2004 Malisoux - 2013 Murray - 2016 Killen - 2011   Murray (2016) compared part of season by training load group, Malisoux (2013) and Ehrmann (2016) only compared injury and pre-injury blocks In some cases (Murray, 2016), a two-way repeated measures ANOVA can examine between and within athlete differences in risk No interactions between multiple predictors Murray (2013) compared load group by season period Assume sphericity Yes, by definition Yes, as season period, or as pre-injury and injury period Cox proportional hazards model 1 √ X X X √ √ √   Malisoux - 2013   Included volume and intensity of training along with age and sex Cox PH conducted at the team/school level examined between-athlete differences No interactions between factors Only included average weekly load and average intensity Can handle unbalanced data By using time-to-event as the outcome, Cox-PH robust to this dependency Uses time to event in analysis 189  Table A.2 (continued): Expanded table of 3-fold alignment (with descriptions)     Themes of theoretical model  Themes of temporal design: intensive longitudinal data Statistical summary and typical uses Method n Multifactorial aetiology Between and within-athlete differences Complex system Includes time-varying and time-invariant variables Missing/unbalanced data (due to rep. measures)  Repeated measure dependency Incorporates time in the analysis Generalised estimating equations  (modeled through logistic and poisson regression) 6 O X X O √ √ O Most GEE approaches accounted for repeated measures, but Clausen et al (2014) used them to cluster players within teams), averaging exposure throughout the entire season.  Clausen - 2014 Cross - 2016 Dennis - 2004 Gabbett - 2010 Gabbett Domrow - 2007 Veugeleurs - 2016   Most authors used multiple variables with GEEs, although some only used GEE to account for repeated measures (Dennis, 2004, Gabbett Domrow, 2007) Provides an 'average' effect for all athletes, but controls for the clustering No complex interactions between factors Some authors only predicted injury (y/n) based on workload variables GEEs handle unbalanced data well  GEE accounts for clustering Do not incorporate time explicitly into the modelling process, and none of the authors included time in the model.  Multilevel Modeling 1 √ √ X √ √ √ X   Windt et al - 2016   Multiple physical outputs included and pre-season training Within-athlete risks determined from level 1 variables, between-athlete from level 2 No complex interactions between factors Pre-season and player variables along with training variables Multilevel models are robust to missing/unbalanced data Via random effects for each player Analysed weekly risk and subsequent week risk; did not directly incorporate time Frailty model  1 √ √ X √ √ √ √ Included a many physical output metrics that were dichotomised into high/low categories and calculated RRs. Multicollinearity was not considered Gabbett Ullah - 2012   Yes, previous injury and physical outputs  Unclear (I think yes due to the frailty term) No complex interactions between factors Yes (injury history) Yes (robust to unbalanced data) Yes (frailty model allows for dependency of recurrent events) Yes, since this is a time-to-event analysis. Appendix B  - Additional tables and figures from Chapter 6 Table B.1: Expanded table of 3-fold alignment (with descriptions) Level of data Purpose in model Variable Scale Units Treatment in modelsa Level 2 Unit of analysis Team    Covariate Average team age Continuous Years Centred on grand mean Covariate UEFA Club Coefficient Continuous Arbitrary Units Centred on grand mean Level 1 Unit of analysis Match observation    Outcome variable (Model 1) Total distance Continuous Meters Raw Outcome variable (Model 2) % of distance covered > high speed (>14 km/h) Continuous Percent Raw Outcome variable (Model 3) Number of sprints Continuous Sprints (count) Raw Independent variable of interest Player unavailability Continuous Players (count) Raw (Random) Covariates Score differential Continuous Goals Raw (Random) Venue Categorical 0 = Away Raw (Random)   1 = Home  Stage Categorical 0 = Group Stage Raw (Random)   1 = Knockout  Ball possession Continuous % of game time Centred within cluster (Random) Level 2 Variables can be included in the model (1) in the raw metric or (2) centred at the grand mean; Level 1 Variables can be included (1) in the raw metric, (2) centred at the grand mean, or (3) centred within the cluster (at the team average). Additionally, Level 1 variables can be fixed (where the effect is assumed to be the same for all teams) or random (where the effect is allowed to vary between teams).   191  Table B.2 – Multilevel linear regression model outputs 2014–2015 season Model 1 — total distance Model 2 — number of sprints Model 3 — percentage of distance covered >14 km/h Fixed effects Coefficient S.E. p-Value Coefficient S.E. p-Value Coefficient S.E. p-Value Intercept β00 113,276.8 706.5  409.1 8.0  28.0 0.25  Player absences β10 596.7 638.3 0.360 9.9 9.6 0.312 0.60 0.27 0.035 Stage (Knockout) β20 −957.4 799.1 0.244 −10.3 11.0 0.363 0 0.3 0.906 Venue (Home) β30 245.9 641.0 0.705 7.6 7.7 0.340 0 0.3 0.858 Score β40 514.7 182.2 0.010 2.5 2.3 0.298 0 0.1 0.511 Ball possession β50 79.0 65.7 0.242 1.5 1.0 0.137 0 0 0.389 UEFA CC β01 −30.7 70.3 0.667 −0.9 0.5 0.115 −0.1 0 0.010 Team age β01 −525.89 420 0.225 1.3 4.4 0.770 0.1 0.1 0.487  Random effects  Standard deviation (p-Value) Standard deviation (p-Value) Standard deviation (p-Value) Intercept τ00 2632.4 (p = 0.022) 24.5 (p = 0.080) 0.8 (p = 0.005) Player absences r10 221.2 (p = 0.246) 6.34 (p = 0.119) 0.2 (p = 0.051) Residual τ11 3122.9 44.2 1.4    192  Table B.2 (continued): Multilevel linear regression model outputs 2015–2016 season Model 1 — total distance Model 2 — number of sprints Model 3 — percentage of distance covered >14 km/h Fixed effects Coefficient S.E. p-Value Coefficient S.E. p-Value Coefficient S.E. p-Value Intercept β00 106,003.5 1250.1  314.0 8.6  25.7 0.35  Player absences β10 1395.3 1085.9 0.214 20.8 9.5 0.040* 0.77 0.32 0.028* Stage (Knockout) β20 2010.6 1209.0 0.113 9.6 11.4 0.411 0.8 0.4 0.089 Venue (Home) β30 589.7 670.2 0.390 17.2 7.4 0.031 0.3 0.3 0.265 Score β40 17.9 219.6 0.936 −0.6 2.1 0.760 −0.1 0.1 0.647 Ball possession β50 −62.4 70.9 0.390 −0.9 0.8 0.262 0 0 0.510 UEFA CC β01 55.1 77.2 0.485 0.8 0.6 0.184 0 0 0.679 Team age β02 −213.1 580.8 0.718 −9.5 3.8 0.023 −0.2 0.2 0.196  Random effects Standard deviation (p-Value) Standard deviation (p-Value) Standard deviation (p-Value) Intercept τ00 4454.5 (p < 0.001) 19.1 (p = 0.015) 0.9 (p = 0.027) Player absences r10 808.3 (p = 0.059) 5.5 (p = 0.036) 0.1 (p = 0.187) Residual τ11 3159.0 34.7 1.3 Bolded rows highlight Player Unavailability Estimates. The estimated ‘average’ effect of player unavailability is the ‘fixed’ effect coefficient, while the variation of this effect across teams is found in the random effects section of the table. *p < 0.05.   193  Table B.3. Team information about injuries and player availability during the 2014–2015/2015–2016 UCL seasons.  2014–2015 2015–2016  Mean ± SD Range Mean ± SD Range Age 25.1 ± 1.3 22.3–27.1 25.4 ± 1.1 23.5–27.7 Height (cm) 182.0 ± 1.53 177.8–185.3 181.7 ± 1.4 178.6–184.3 Weight (kg) 77.8 ± 2.3 73.1–82.9 77.6 ± 1.6 73.3–79.4 Squad size (n) 31.2 ± 5.0 23–40 31.0 ± 5.2 24–41 Injury totals (n)      - Overall 41.2 ± 14.1 17–70 38.7 ± 12.8 19–66  - Match 23.3 ± 9.2 11–45 22.5 ± 8.0 8–39  - Training 17.9 ± 8.0 3–40 16.2 ± 7.7 3–32 Injury incidence (/1000 h)      - Overall 5.7 ± 2.3 1.6–10.8 6.1 ± 2.4 2.9–12.8  - Match 20.6 ± 7.9 8.8–40.5 20.8 ± 6.8 8.9–34.0  - Training 3.0 ± 1.3 0.3–6.3 3.1 ± 1.5 0.43–6.9 Injury type      - Acute 26.2 ± 10.3 9–42 23.6 ± 11.3 6–45  - Overuse 15.0 ± 8.9 4–38 15.1 ± 7.6 4–32    194   Figure B.1: The average number of players unavailable due to injury for games in each month of competition play. Note — there were no matches played in January.    195  Appendix C: Note – thanks to Carson Grose and David Taylor for assistance with this proof.   𝑼𝒏𝒄𝒐𝒖𝒑𝒍𝒆𝒅 =  𝑾𝟏𝟎.𝟑𝟑(𝑾𝟐+𝑾𝟑+𝑾𝟒)                                            𝑪𝒐𝒖𝒑𝒍𝒆𝒅 =  𝑾𝟏𝟎.𝟐𝟓(𝑾𝟏+𝑾𝟐+𝑾𝟑+𝑾𝟒)  𝑾𝟏 =  𝑨𝒄𝒖𝒕𝒆 𝝁 =  𝑾𝟐 + 𝑾𝟑 + 𝑾𝟒 Therefore;  𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =   𝐴0.25(𝐴 + 𝜇)=  4𝐴(𝐴 + 𝜇)  𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 =   𝐴0.33(𝜇)=  3𝐴(𝜇) 𝐴 =   𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇3 Therefore; 𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =   (4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇3)  ÷ (𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇3+ 𝜇) 𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =   (4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇3)  ÷ (𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇 + 3𝜇3) 𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =  4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇 + 3𝜇  𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =  4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑. 𝜇 𝜇(𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 + 3) 𝑪𝒐𝒖𝒑𝒍𝒆𝒅 =  𝟒. 𝑼𝒏𝒄𝒐𝒖𝒑𝒍𝒆𝒅 𝑼𝒏𝒄𝒐𝒖𝒑𝒍𝒆𝒅 + 𝟑  𝐶𝑜𝑢𝑝𝑙𝑒𝑑(𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 + 3) =  4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 𝐶𝑜𝑢𝑝𝑙𝑒𝑑. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 + 3. 𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =  4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 3. 𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =  4. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 − 𝐶𝑜𝑢𝑝𝑙𝑒𝑑. 𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑 3. 𝐶𝑜𝑢𝑝𝑙𝑒𝑑 =  𝑈𝑛𝑐𝑜𝑢𝑝𝑙𝑒𝑑(4 − 𝐶𝑜𝑢𝑝𝑙𝑒𝑑) 𝑼𝒏𝒄𝒐𝒖𝒑𝒍𝒆𝒅 =  𝟑. 𝑪𝒐𝒖𝒑𝒍𝒆𝒅 𝟒 − 𝑪𝒐𝒖𝒑𝒍𝒆𝒅 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items