The impact of different training load quantification and modelling methodologies on performance predictions in elite swimmers

Validity and Accuracy of Impulse-Response Models for Modeling and Predicting Training Effects on Performance of Swimmers

PURPOSE

The aim of this study was to compare the suitability of models for practical applications in training planning.

METHODS

We tested six impulse-response models, including Banister's model (Model Ba), a variable dose-response model (Model Bu), and indirect-response models differing in the way they account or not for the effect of previous training on the ability to respond effectively to a given session. Data from 11 swimmers were collected during 61 wk across two competitive seasons. Daily training load was calculated from the number of pool-kilometers and dry land workout equivalents, weighted according to intensity. Performance was determined from 50-m trials done during training sessions twice a week. Models were ranked on the base of Aikaike's information criterion along with measures of goodness of fit.

RESULTS

Models Ba and Bu gave the greatest Akaike weights, 0.339 ± 0.254 and 0.360 ± 0.296, respectively. Their estimates were used to determine the evolution of performance over time after a training session and the optimal characteristics of taper. The data of the first 20 wk were used to train these two models and predict performance for the after 8 wk (validation data set 1) and for the following season (validation data set 2). The mean absolute percentage error between real and predicted performance using Model Ba was 2.02% ± 0.65% and 2.69% ± 1.23% for validation data sets 1 and 2, respectively, and 2.17% ± 0.65% and 2.56% ± 0.79% with Model Bu.

CONCLUSIONS

The findings showed that although the two top-ranked models gave relevant approximations of the relationship between training and performance, their ability to predict future performance from past data was not satisfactory for individual training planning.

Training Load and Acute Performance Decrements Following Different Training Sessions

PURPOSE

To examine the differences in training load (TL) metrics when quantifying training sessions differing in intensity and duration. The relationship between the TL metrics and the acute performance decrement measured immediately after the sessions was also assessed.

METHODS

Eleven male recreational cyclists performed 4 training sessions in a random order, immediately followed by a 3-km time trial (TT). Before this period, participants performed the time TT in order to obtain a baseline performance. The difference in the average power output for the TTs following the training sessions was then expressed relative to the best baseline performance. The training sessions were quantified using 7 different TL metrics, 4 using heart rate as input, 2 using power output, and 1 using the rating of perceived exertion.

RESULTS

The load of the sessions was estimated differently depending on the TL metrics used. Also, within the metrics using the same input (heart rate and power), differences were found. TL using the rating of perceived exertion was the only metric showing a response that was consistent with the acute performance decrements found for the different training sessions. The Training Stress Score and the individualized training impulse demonstrated similar patterns but overexpressed the intensity of the training sessions. The total work done resulted in an overrepresentation of the duration of training.

CONCLUSION

TL metrics provide dissimilar results as to which training sessions have higher loads. The load based on TL using the rating of perceived exertion was the only one in line with the acute performance decrements found in this study.

The Current State of Subjective Training Load Monitoring: Follow-Up and Future Directions

This article addresses several key issues that have been raised related to subjective training load (TL) monitoring. These key issues include how TL is calculated if subjective TL can be used to model sports performance and where subjective TL monitoring fits into an overall decision-making framework for practitioners. Regarding how TL is calculated, there is conjecture over the most appropriate (1) acute and chronic period lengths, (2) smoothing methods for TL data and (3) change in TL measures (e.g., training stress balance (TSB), differential load, acute-to-chronic workload ratio). Variable selection procedures with measures of model-fit, like the Akaike Information Criterion, are suggested as a potential answer to these calculation issues with examples provided using datasets from two different groups of elite athletes prior to and during competition at the 2016 Olympic Games. Regarding using subjective TL to model sports performance, further examples using linear mixed models and the previously mentioned datasets are provided to illustrate possible practical interpretations of model results for coaches (e.g., ensuring TSB increases during a taper for improved performance). An overall decision-making framework for determining training interventions is also provided with context given to where subjective TL measures may fit within this framework and the determination if subjective measures are needed with TL monitoring for different sporting situations. Lastly, relevant practical recommendations (e.g., using validated scales and training coaches and athletes in their use) are provided to ensure subjective TL monitoring is used as effectively as possible along with recommendations for future research.

The Fitness-Fatigue Model: What's in the Numbers?

PURPOSE

The purpose of this commentary is to outline some of the pitfalls when using the fitness-fatigue model to unravel the interaction between training load and performance. By doing so, we encourage sport scientists and coaches to interpret the parameters from the model with some extra caution.

CONCLUSIONS

Caution is needed when interpreting the fitness-fatigue model since the parameter values are influenced by the starting parameter values, the modeling technique, and the input of the model. Also, the use of general constants should be avoided since they do not account for interindividual differences and differences between training-load methods. Therefore, we advise sport scientists and coaches to use the model as a way to work more data-informed rather than working data-driven.

The Use of Fitness-Fatigue Models for Sport Performance Modelling: Conceptual Issues and Contributions from Machine-Learning

The emergence of the first Fitness-Fatigue impulse responses models (FFMs) have allowed the sport science community to investigate relationships between the effects of training and performance. In the models, athletic performance is described by first order transfer functions which represent Fitness and Fatigue antagonistic responses to training. On this basis, the mathematical structure allows for a precise determination of optimal sequence of training doses that would enhance the greatest athletic performance, at a given time point. Despite several improvement of FFMs and still being widely used nowadays, their efficiency for describing as well as for predicting a sport performance remains mitigated. The main causes may be attributed to a simplification of physiological processes involved by exercise which the model relies on, as well as a univariate consideration of factors responsible for an athletic performance. In this context, machine-learning perspectives appear to be valuable for sport performance modelling purposes. Weaknesses of FFMs may be surpassed by embedding physiological representation of training effects into non-linear and multivariate learning algorithms. Thus, ensemble learning methods may benefit from a combination of individual responses based on physiological knowledge within supervised machine-learning algorithms for a better prediction of athletic performance.In conclusion, the machine-learning approach is not an alternative to FFMs, but rather a way to take advantage of models based on physiological assumptions within powerful machine-learning models.

The Impact of a Swimming Training Season on Anthropometrics, Maturation, and Kinematics in 12-Year-Old and Under Age-Group Swimmers: A Network Analysis

Understanding fluctuations and associations between swimming performance-related variables provide strategic insights into a swimmer's preparation program. Through network analysis, we verified the relationships between anthropometrics, maturation, and kinematics changes (Δ) in 25-m breaststroke (BREAST) and butterfly (FLY) swimming performance, before and after a 47-week swimming training season. Twenty age-group swimmers (n =11 girls: 10.0 ± 1.3 years and n = 9 boys: 10.5 ± 0.9 years) performed a 25-m all-out swim test (T25) in BREAST and FLY techniques, before and after 47 weeks. Three measures of centrality, transformed into a z-score, were generated: betweenness, closeness, and strength. Data were compared (t-test) and effect sizes were identified with Hedges' g. Large effect sizes were observed for swimming performance improvements in BREAST (32.0 ± 7.5 to 24.5 ± 3.8 s; g = 1.26; Δ = −21.9 %) and FLY (30.3 ± 7.0 to 21.8 ± 3.6 s; g = 1.52; Δ = −26.5 %). Small to moderate effect sizes were observed for anthropometric changes. Moderate effect size was observed for maturity offset changes (−2.0 ± 0.9 to −1.3 ± 1.0; g = 0.73; Δ = 50.9 ± 281 %). Changes in maturity offset, stroke rate (SR), and stroke length for both BREAST and FLY swimming speeds were highlighted by the weight matrix. For betweenness, closeness, and strength, changes in arm span (AS) (BREAST) and stroke length (FLY) were remarkable. The dynamic process of athletic development and the perception of complexity of fluctuations and associations between performance-related variables were underpinned, particularly for simultaneous swimming techniques in age-group swimmers.

Training load responses modelling and model generalisation in elite sports

This study aims to provide a transferable methodology in the context of sport performance modelling, with a special focus to the generalisation of models. Data were collected from seven elite Short track speed skaters over a three months training period. In order to account for training load accumulation over sessions, cumulative responses to training were modelled by impulse, serial and bi-exponential responses functions. The variable dose-response (DR) model was compared to elastic net (ENET), principal component regression (PCR) and random forest (RF) models, while using cross-validation within a time-series framework. ENET, PCR and RF models were fitted either individually (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{I}$$\end{document}MI) or on the whole group of athletes (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{G}$$\end{document}MG). Root mean square error criterion was used to assess performances of models. ENET and PCR models provided a significant greater generalisation ability than the DR model (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = 0.018$$\end{document}p=0.018, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p < 0.001$$\end{document}p<0.001, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = 0.004$$\end{document}p=0.004 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p < 0.001$$\end{document}p<0.001 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ENET_{I}$$\end{document}ENETI, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ENET_{G}$$\end{document}ENETG, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$PCR_{I}$$\end{document}PCRI and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$PCR_{G}$$\end{document}PCRG, respectively). Only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ENET_{G}$$\end{document}ENETG and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$RF_{G}$$\end{document}RFG were significantly more accurate in prediction than DR (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p < 0.001$$\end{document}p<0.001 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p < 0.012$$\end{document}p<0.012). In conclusion, ENET achieved greater generalisation and predictive accuracy performances. Thus, building and evaluating models within a generalisation enhancing procedure is a prerequisite for any predictive modelling.

International survey of training load monitoring practices in competitive swimming: How, what and why not?

OBJECTIVE

The purpose of this study is to identify the training load (TL) monitoring practices employed in real-world competitive swimming environments. The study explores data collection, analysis and barriers to TL monitoring.

DESIGN

Cross-sectional.

SETTING

Online survey platform.

PARTICIPANTS

Thirty-one responders working in competitive swimming programmes.

MAIN OUTCOME MEASURES

Methods of data collection, analysis, level of effectiveness and barriers associated with TL monitoring.

RESULTS

84% of responders acknowledged using TL monitoring, with 81% of responders using a combination of both internal and external TL, in line with current consensus statements. Swim volume (mileage) (96%) and session rate of perceived exertion (sRPE) (92%) were the most frequently used, with athlete lifestyle/wellness monitoring also featuring prominently. Thematic analysis highlighted that "stakeholder engagement", "resource constraints" or "functionality and usability of the systems" were shared barriers to TL monitoring amongst responders.

CONCLUSIONS

Findings show there is a research-practice gap. Future approaches to TL monitoring in competitive swimming should focus on selecting methods that allow the same TL monitoring system to be used across the whole programme, (pool-based training, dryland training and competition). Barriers associated with athlete adherence and coach/National Governing Body engagement should be addressed before a TL system implementation.

Predicting performance in 4 x 200-m freestyle swimming relay events

AIM

The aim was to predict and understand variations in swimmer performance between individual and relay events, and develop a predictive model for the 4x200-m swimming freestyle relay event to help inform team selection and strategy.

DATA AND METHODS

Race data for 716 relay finals (4 x 200-m freestyle) from 14 international competitions between 2010-2018 were analysed. Individual 200-m freestyle season best time for the same year was located for each swimmer. Linear regression and machine learning was applied to 4 x 200-m swimming freestyle relay events.

RESULTS

Compared to the individual event, the lowest ranked swimmer in the team (-0.62 s, CI = [-0.94, -0.30]) and American swimmers (-0.48 s [-0.89, -0.08]) typically swam faster 200-m times in relay events. Random forest models predicted gold, silver, bronze and non-medal with 100%, up to 41%, up to 63%, and 93% sensitivity, respectively.

DISCUSSION

Team finishing position was strongly associated with the differential time to the fastest team (mean decrease in Gini (MDG) when this variable was omitted = 31.3), world rankings of team members (average ranking MDG of 18.9), and the order of swimmers (MDG = 6.9). Differential times are based on the sum of individual swimmer's season's best times, and along with world rankings, reflect team strength. In contrast, the order of swimmers reflects strategy. This type of analysis could assist coaches and support staff in selecting swimmers and team orders for relay events to enhance the likelihood of success.

The Training-Performance Puzzle: How Can the Past Inform Future Training Directions?

Over the past 20 years, research on the training-load-injury relationship has grown exponentially. With the benefit of more data, our understanding of the training-performance puzzle has improved. What were we thinking 20 years ago, and how has our thinking changed over time? Although early investigators attributed overuse injuries to excessive training loads, it has become clear that rapid spikes in training load, above what an athlete is accustomed, explain (at least in part) a large proportion of injuries. In this respect, it appears that overuse injuries may arise from athletes being underprepared for the load they are about to perform. However, a question of interest to both athletic trainers (ATs) and researchers is why some athletes sustain injury at low training loads, while others can tolerate much greater training loads? A higher chronic training load and well-developed aerobic fitness and lower body strength appear to moderate the training-injury relationship and provide a protective effect against spikes in load. The training-performance puzzle is complex and dynamic-at any given time, multiple inputs to injury and performance exist. The challenge facing researchers is obtaining large enough longitudinal data sets to capture the time-varying nature of physiological and musculoskeletal capacities and training-load data to adequately inform injury-prevention efforts. The training-performance puzzle can be solved, but it will take collaboration between researchers and clinicians as well as an understanding that efficacy (ie, how training load affects performance and injury in an idealized or controlled setting) does not equate to effectiveness (ie, how training load affects performance and injury in the real-world setting, where many variables cannot be controlled).