A comparison of the most intense periods (MIPs) during competitive matches and training over an 8-week period in a male elite �eld hockey team

Purpose Wearables serve to quantify the on-court activity in intermittent sports such as �eld hockey (FH). Based on the objective data, benchmarks can be elaborated to tailor training intensity and volume. Next to average and accumulated values, the most intense periods (MIPs) during competitive FH matches are from special interest, since these quantify the peak intensities players experience throughout the intermittent matches. The aim of this study was to retrospectively compare peak intensities between training and competition sessions in a male FH team competing in the �rst german division. Methods Throughout an 8-week in-season period


Introduction
In high-performance sports, analyzing an athlete´s activity data has become crucial for gaining insights into their physical and physiological demands during competition.This information is essential for optimizing performance and improving overall athletic abilities (Bourdon et al., 2017).Activity data quanti es the athletes´ work during competition and can be categorized into external and internal load data.While the external load describes the physical work of the athletes like distances covered or velocities reached, the internal load describes the individual physiological responses to the given external load such as heart rate increases or lactate accumulation (Halson, 2014).Nowadays, wearable technologies such as global positioning system (GPS) sensors and accelerometers allow easy, automatic, and precise recording of external load data to describe athlete activity pro les (Torres-Ronda et al., 2022).
Based on this objective quanti cation of the physical work of the athlete on the eld, benchmarks can be developed to individualize the intensity and duration of exercise during exercise sessions (Delaney et al., 2017).These benchmarks allow for a tailored adaptation of training and aim to optimize the preparation of the athlete for the anticipated physical stress during competition (Delaney et al., 2017).
In team sports such as eld hockey (FH), the intermittent and unpredictable nature of sports challenges the elaboration of benchmarks, since external loads are highly variable and dynamic.Therefore, a FH player is not only required to sustain physical performance over time (4 x 15 minutes), but also to follow the unpredictable increments and reductions in match intensity through a FH match.The overall demands of a player can be quanti ed due to the accumulation of recorded activity data.Throughout four quarters of 15 minutes each, FH players cover 6 to 8 km per match at an average speed of 7.5 to 9 km/h (Sunderland & Edwards, 2017).From this total distance, ~ 10% is covered at intense running speeds > 16 km/h, while > 50% of the distance is covered by jogging and walking, so there is high variability in the intensity of activity (McGuinness et al., 2018).One limitation of accumulative analysis of activity data is that it barely re ects the intermittent and variable character of activity pro les in FH.To overcome this issue, more recent data analysis approaches investigated activity data within shorter time windows to identify the most intense periods (MIPs) within the intermittent activity pro les (Cunningham et al., 2018).These MIPs are further stated as worst-case scenarios since they display the upper limit of physiological demands that an athlete can experience during a match (Weaving et al., 2022).The MIP approach is less sensitive to variations in competition intensities and avoids that activity peaks being washed out by FH-speci c phases of inactivity like penalties or time-outs (Whitehead et al., 2018).
Therefore, the MIP approach provides complementary information on the on-eld demands of FH players by re ecting the potential extremes of on-eld demands.Therefore, the present study aimed to compare the MIPs extracted from training and competitive sessions in a high-level male FH team.To identify the potential discrepancies between peak intensities during training and competition, an 8-week competitive in-season cycle was investigated retrospectively.Based on previous research (Duthie et al., 2019;Oliva-Lozano et al., 2022), it was hypothesized that MIPs from competition exceed those extracted from training sessions.In order to take the intermittent character of FH into account, this study further not only explored distance-related but also acceleration-related outcomes as an important cornerstone of FH performance (Duthie et al., 2019).Upon identi cation of discrepancies in MIPs between competition and training, this study may provide important implications for the design of training sessions according to actual competition-derived peak intensities.

Sample
In total, data sets from 20 elite FH players (21.2 ± 2.4 years) were collected throughout the 8-week inseason cycle.All players were part of a german eld hockey team competing in the rst german division.
The players were amateurs, trained three times a week, and competed in one to two matches on the weekend.Further, all players performed individual strength and conditioning sessions besides their regular training and profession/ education.The players were informed about the analysis of the data collected throughout the training and competition since the data was embedded into the training monitoring process of the team.Through written informed consent, every player agreed to analyze his data for scienti c purposes.Further, the ethics committee of the a liated university approved the analysis of the training data.

Training cycle
The 8-week period analyzed in the present study resembles the rst competitive part of the season.During this period, the team trained three times a week (excluding individual strength and conditioning sessions) and played one to two competitive matches on weekends.In total, 30 sessions were recorded throughout the cycle (11 matches), resulting in a total number of n = = 373 individual raw data sets.The weekly training cycle was organized into an intense training session on Tuesday (T1), a technical training session on Wednesdays with a focus on team tactics and corner shooting (T2), and a game-orientated training session on Tuesday (T3).In three of the eight weeks, two competitive matches (C) had to be played on one weekend.

Extraction of MIPs
Running and acceleration data was recorded using the Polar Team Pro sensor (Polar Electronics, Kempele, Finland).The sensor was connected to a chest belt at the beginning of each session and the recording session was started at the beginning of the warm-up period.The sensor records positional data at 10 Hz due to a GPS sensor and further records acceleration data at 200 Hz due to an inertial measurement unit (IMU).The system has been used previously to analyze activity data in outdoor team sports (Reinhardt et al., 2019).For the determination of peak match demands, raw data including sample points, accumulative distance (in m), acute speed (in km•h − 1 ) and acceleration (in m•s − 2 ) was exported from the Polar Team Pro cloud at a sampling rate of 10 Hz.To extract the MIPs for each session throughout the training cycle, the raw data for each session was imported into MATLAB (R2020b, The MathWorks) and analyzed based on original scripts.The MIPs were calculated by applying the rolling window approach since this approach outperformed the xed period method to identify peak intensities (Cunningham et al. 2018).Following Delves et al (2019), we chose to roll window lengths of 1, 2, 3, 4, and 5 minutes to calculate MIPs.Since the raw data displayed the accumulated distance at 10 Hz, the algorithm rst calculated sample-to-sample changes and then computed the maximum moving sum over the above-mentioned window lengths (1-5 minutes) as session MIP.For high-intensity running (HIR; > 16 km•h − 1 ), a similar approach was chosen, but only sample-to-sample changes resulting from measured acute speeds higher than 16 km•h − 1 were considered.For sprinting (SPRINT; > 20 km•h − 1 ), only sampleto-sample changes resulting from measured acute speeds < 20 km•h − 1 were considered.Then, sample-tosample changes were computed before the computation of moving sums over the above-speci ed window lengths.For the acceleration metrics, both negative and positive acceleration values were turned into absolute values, changing negative into non-negative values, and moving averages over the abovespeci ed window length were calculated over the whole time series (Delaney et al., 2016).
For training sessions, the whole session was analyzed starting with the warm-up and ending after the cool-down activity.For matches, the recorded session was cut into two sections (section 1: quarter 1 + quarter 2, section 2: quarter 3 + quarter 4) and each section was analyzed as a single session for MIP detection.In the nal step, the higher MIP of both sections was identi ed as the competition MIP.MIPs served as dependent variables and were calculated in m•min − 1 for total distance (TD), HIT distance

Statistics
All statistical analyses were performed using MATLAB R2020a (Version 29, IBM, New York).First, the data were checked for normal distribution by applying the Kolmogorov-Smirnov test.Since the majority of outcomes under the independent variable "training" displayed a non-parametric distribution, Mann-Whitney-Tests were applied to analyze differences between training and competition MIPs.The statistical level of signi cance for all tests was set to p < .05.To interpret training-competition discrepancies between different MIP parameters and window lengths, effect sizes were calculated according to Cohen´s d.

Descriptive absolute data
Across the 8-week in-season cycle, 30 sessions (11 matches) were recorded and resulted in 372 individual activity data recordings (C = 144, T1 = 56, T2 = 84, T3 = 88).The average total distance, HRUN distance, and SPRINT distance covered were signi cantly higher during the competition (p < .001).An overview of average values from training and competitive sessions is shown in Table 1.

Mips
To assess differences in MIPs between competitive matches and training sessions, Mann-Whitney-U-Tests were conducted.The analysis revealed highly signi cant differences between MIPs extracted from competition and training sessions for the metrics TD, ACCEL, HIR, and SPRINT (p < .001).An overview of p-statistics and corresponding effect sizes is presented in Table 1.High effect sizes were observed for TD, d = 1.14 to 1.35), HIR (d = 1.2 to 2.31), and SPRINT (d = 1.22 to 1.89).For ACCEL, moderate to high effect sizes were observed (d = .49to .92).
Outcome-dependent trends have been observed for the associations between effect size and MIP window length across parameters.While effect sizes for ACCEL increased with longer MIP window length, effect sizes tend to decrease with longer window length for SPRINT MIPs.For TD and HIR, no speci c trends were observed.Figures 1 and 2

Conclusions
The present study objectively identi ed discrepancies in peak intensities between training and competitive sessions in a male rst-division eld hockey team.The discrepancies were evident for both distance-as well as acceleration-based metrics and are likely to be caused by training routines related to SSGs.Therefore, coaches and scientists should consider an evaluation of actual match peak intensities as benchmarks for the adjustment of training routines.Next to peak intensities, aspects such as volume should be further considered to optimally prepare athletes for competing demands.Future studies should MIPs from competitive FH matches and observed considerable differences between average match intensities (about 120 m•min − 1 ) and peak match intensities based on the MIP approach (> 200 m•min − 1 ) (Delves et al., 2019; McGuinness et al., 2020).Consequently, training intensities adapted according to accumulated values would be likely to underestimate the peaks of intensity during competitive matches (Weaving et al., 2022).In particular, a comparison of the training load during 5-minute peak intervals in matches and 5 minutes of small-sided games (SSGs) during training indicated that the latter only partially reached the intensities of game demands (Duthie et al., 2019).While all studied SSG variations (2vs2, 3vs3, 4vs4) demonstrated higher overall acceleration values compared to the competition, substantially lower values were observed for distance-based metrics (Duthie 2019).A retrospective analysis in soccer came to similar conclusions after demonstrating that training sessions failed to reproduce MIPs from competitions in various distance-based MIP metrics such as total distance covered or high-intensity running distance (Oliva-Lozano et al., 2022).Aiming to prepare athletes for competitive demands, it seems likely that training sessions in team sports do not reach the peak intensities players experience during matches.Therefore, the extraction of MIPs from competitions and training can serve as valuable input to re-calibrate training practices and intensities.
, and SPRINT distance (m•min − 1 > 20 km•h − 1 ).The MIPs for the acceleration load (ACCEL) were calculated in m•s − 2 •min − 1 .The session type (competition vs. training [T1 & T3]) served as the independent variable.Due to the focus on technical-tactical training content, T2 sessions were excluded from the statistical comparison.

Figures Figure 1
Figures

Table 1
further display histogram plots of the MIPs extracted from training and competition.Overview of average and SD values of different external load metrics extracted from training and competitive sessions throughout an 8-week in-season period in high-level male FH Parameters resemble total distance high-intensity-running distance (HIR, [> 16 km/h]), sprinting distance (SPRINT, > 20 km/h), and acceleration (ACCEL).Most intense periods (MIPs) were extracted over different window lengths ranging from 1 minute to 5 minutes.As a key nding of this study, the analysis revealed that distance-based MIPs during training signi cantly failed to reach the magnitude of peak intensities observed during competitive matches.As shown in Figs. 1 and 2, TD-, HIR-and SPRINT-based MIPs did not reach the intensities the players experienced during competitive matches on the weekend.In comparison to a previous study extracting MIPs from matches in high-level female FH players by McGuinness et al. (2022) and Delves (2019), the MIPs extracted in the present study were slightly lower, particularly for high-intensity MIPs (HIR and SPRINT).However, the MIPs experienced during the competition were still ~ 40 to 60% higher than the peak intensities observed during the training sessions.Therefore, it might be stated that the intensity of training sessions was not high enough to prepare the players for competition intensities.loadonthe players compared to training sessions.Similar observations have been made in professional handball players (Font et al., 2022) and youth soccer players (Szigeti et al., 2021) through retrospective analyses.On the one hand, a tapering of training load between matches seems reasonable to reduce the metabolic and mechanical load and boost the performance capacities of a player (Vachon et al., 2021).However, exercise sessions that are not representative of competitive characteristics are likely to fail to induce the physiological overload needed for physiological adaptations relevant to performance (Delaney et al., 2017).The present study identi ed a discrepancy between training and matches also for the maximum intensities experienced during exercise.One issue of not preparing players for competitive MIPs during training is that the latter is associated with fatigue-initiated reductions in physical performance.Schimpchen et al. (2021) observed reduced high-intensity running activity for ve consecutive minutes after individual MIPs.Therefore, MIPs seem to display severe physiological stressors athletes are not familiarized with during regular exercise sessions.A possible reason for this discrepancy between peak intensities could be found in training routines in team sports, such as the focus on technical-tactical content during the season and the utilization of SSGs to simulate game situations (Weaving et al., 2022).For instance, Duthie et al. (2019) already reported in a sample of FH players that typical training contents, e.g.SSGs, do not reach the distance-related intensity of competitive matches, but do so for acceleration-based metrics.Lacome et al. (2018) also described this phenomenon concerning MIPs in soccer players comparing SSGs and competitive matches (Lacome et al., 2018).Among other reasons, the reduction of court dimensions and potential running distances was discussed as a possible reason for lower distance-based intensities.Findings from Martin-Garcia (2019) on soccer players support the observation that particularly for high-speed activities, SSGs seem insu cient to stimulate competition-like intensities.Further, the ecological context of training drills seems to affect the peak intensities a player can experience.For instance, Sansone et al. (2023) reported that the focus on skill-related exercises exposes players to lower intensities than game-based exercises (Sansone et al., 2022).Therefore, the present study supports previous studies claiming insu cient intensities in (FH?) team sports training, with a speci c focus on MIPs as short, but strenuous periods of activity.A possible solution could be increasing the court dimensions in the SSG to maintain the spatial characteristics of the team sports and avoid reductions in speed and running demands (Lacome et al., 2018).In addition to distance-based metrics, acceleration-based MIPs revealed discrepancies between training and competition intensities.Particularly for MIPs calculated across time windows ≥ 2 minutes, signi cantly higher MIPs were observed for competitive sessions.Based on effect size estimation and the observation of MIP distribution in Fig. 1, it can be stated that training-competition discrepancies for acceleration-based MIPs are less drastic than for distance-based metrics (TD, HIR, and SPRINT).Furthermore, the effect sizes indicate higher differences between competition and training for a larger MIP window length.Considering that several studies suggested that SSGs better match accelerationthan distance-based competition intensities and partly extend the latter (Duthie et al., 2019; Lacome et al., 2018; Oliva-Lozano et al., 2022), the reduction of court dimensions seems to be effective in preserving game-like acceleration demands.However, the signi cant differences in acceleration data observed in the present study are likely to result from the pooling of all recorded exercise sessions within the analyzed period of the season.Considering the variability of training intensities in terms of micro-cycle periodization, it seems legit that not all exercise sessions reach competition intensity (Oliva-Lozano et al., 2020, 2022).Furthermore, the duration of SSGs is frequently restricted to bouts of a few minutes with intermittent breaks (Duthie et al., 2019), so an increased intensity over several minutes cannot be reached in such types of game-based exercise.As a nal observation of this study, it is worth noting that the discrepancy between training and competition was in parts moderated by the window length chosen for MIP extraction.Several studies indicated that the length of the analyzed window modulates the severity of differences between the average and peak intensity and highlighted the importance of the choice of window length for the description of peak intensities (Delves et al., 2019; McGuinness et al., 2020).For comparison of training and competition data, the length of the chosen window had a substantial impact on ACCEL-based MIPs (Table 1) and ampli ed the discrepancy between training and competition.Further, the different metabolic demands associated with high-intensity exercise such as SSGs of varying window length (e.g. 1 minute vs. 5 minutes) should be considered when using objective MIP values as benchmarks for training design (Gaudino et al., 2014).To prepare an athlete for the intermittent and variable demands of the competition, it could be important to implement SSG formats of varying lengths for simulating worstcase scenarios experienced during competition across different time scales.LimitationsDespite the important implications of this study for exercise planning, several potential limitations must be considered.First, only one team was analyzed, so team-speci c routines may have contributed to the discrepancy between training and competition MIPs.Although such discrepancies have been reported previously (Lacome et al., 2018), it is highly recommended to compare MIPs from training and competition for each team speci cally before re-calibrating training intensities.Further, the present approach is restricted to the physical demands of team sports.While training routines such as SSGs seem to not match the physical requirement of the competition yet, an intensi cation of non-physical aspects such as tactics and techniques compared to the match demands might be still given (Clemente et al., 2021).Therefore, coaches and scientists are requested to treat the observation of the present study as a piece of complementary information for the optimization of training.
DiscussionThe present study aimed to retrospectively compare MIPs extracted from training and competitive sessions throughout an 8-week in-season cycle in an elite male FH team.The analysis of more than 300 individual recordings revealed that peak intensities during training neither reached competition intensities for distance-nor acceleration-based metrics.A comparison of effect sizes between the different inspected outcomes (TD, HIR, SPRINT, and ACCEL) indicated higher discrepancies between training and competition for distance-based metrics.No systematic differences were observed concerning the length of the window chosen for MIP extraction.The observed discrepancies between training and competition intensities provide important implications for the design of training sessions in FH and team sports.Discrepancies in intensities between training and competitive session intensity have been observed and reported in previous analyses of team sports.For instance, Oliva-Lozano et al. (2022) analyzed activity data from professional soccer players over a consecutive in-season period and reported that match days induced a higher