Content uploaded by Ermanno Rampinini
Author content
All content in this area was uploaded by Ermanno Rampinini on Feb 02, 2018
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=rsmf20
Science and Medicine in Football
ISSN: 2473-3938 (Print) 2473-4446 (Online) Journal homepage: http://www.tandfonline.com/loi/rsmf20
Despite association, the acute:chronic work load
ratio does not predict non-contact injury in elite
footballers
Maurizio Fanchini, Ermanno Rampinini, Marco Riggio, Aaron J. Coutts,
Claudio Pecci & Alan McCall
To cite this article: Maurizio Fanchini, Ermanno Rampinini, Marco Riggio, Aaron J. Coutts,
Claudio Pecci & Alan McCall (2018): Despite association, the acute:chronic work load ratio
does not predict non-contact injury in elite footballers, Science and Medicine in Football, DOI:
10.1080/24733938.2018.1429014
To link to this article: https://doi.org/10.1080/24733938.2018.1429014
Published online: 24 Jan 2018.
Submit your article to this journal
View related articles
View Crossmark data
ARTICLE
Despite association, the acute:chronic work load ratio does not predict non-contact
injury in elite footballers
Maurizio Fanchini
a,b
, Ermanno Rampinini
c
, Marco Riggio
a
, Aaron J. Coutts
d
, Claudio Pecci
c
and Alan McCall
e,f
a
US Sassuolo Football Club, Sassuolo, Italy;
b
Department of Neurosciences, Biomedicine and Movement Sciences, University of Verona, Verona,
Italy;
c
MAPEI Sport Research Centre, Olgiate Olona, Italy;
d
Sport and Exercise Discipline Group, University Technology Sydney (UTS), Moore Park,
NSW, Australia;
e
Research & Development Department, Arsenal Football Club, London, UK;
f
Research & Development Department, Edinburgh
Napier University, Edinburgh, UK
ABSTRACT
Purpose: To examine association and prediction of load-based markers (rate of perceived exertion) with
non-contact injuries.
Materials and methods: Thirty-four elite Italian football players (age 26 ± 5 y, height 182 ± 5 cm, body
mass 78 ± 4 kg) participated in a 3-seasons prospective study. Markers examined were: RPE, exposure,
weekly load, week-to-week load change, cumulative 2, 3, 4 WL, acute:chronic 1:2 (acute:chronic2), 1:3
(acute:chronic3) and 1:4 (acute:chronic4) WL ratios. After checking multicollinearity between markers, a
Generalized Estimating Equation analysis was used to examine association with a non-contact injury in
the subsequent week. The associated markers were split into four groups based on 15th, 50th, 85th
percentile to compare injury risk (IR) in different zones. Prediction was examined with receiver operat-
ing characteristic curve, area under the curve (AUC) and Youden index.
Results: IR increased when acute:chronic2 of 1.00–1.20, >1.20 were compared to <0.81 (odds ratio (OR),
90% confidence interval (CI): 1.6, 0.79–3.29; 2.2, 1.03–4.74). IR increased when comparing acute:chronic3
of 1.01–1.23, >1.23 vs. <0.80 (OR, 90% CI: 1.9, 0.9–3.8; 2.5, 1.2–5.4). IR increased when comparing acute:
chronic4 of 0.78–1.02, 1.02–1.26, >1.26 vs. <0.78 (OR, 90% CI: 2.4, 1.4–3.9; 3.3, 1.6–6.6; 3.5, 1.7–7.4). The
AUC ≤0.60 for all markers and Youden Index (close to 0) showed poor prediction.
Conclusion: Acute:chronic markers showed association however with poor prediction ability.
ARTICLE HISTORY
Accepted 10 January 2018
KEYWORDS
Team sport; soccer; injury
prevention; training load
Introduction
The relationship between training load and injury in ath-
letes has recently received increased attention in research
and practical settings. Indeed, the International Olympic
Committee (IOC) recently released a consensus statement
(Soligard et al. 2016) highlighting that the relative load (i.e.,
the rate of load application relative to what the athlete has
been prepared for) is the most pertinent measure of load
associated with injury in athletes. Referred to as the “acute:
chronic ratio”(acute:chronic) (typically, acute = current
week and chronic = rolling 4 week average), this measure
has been suggested to reflect the ratio between fatigue (i.e.,
acute) and fitness (i.e., chronic) of the player and has been
associated with injury in many sports (Malone, Owen, et al.
2016;Hulinetal.2014,2016a;Malone,Roeetal.2016;
Murray et al. 2016; Weiss et al. 2017). In general, it is
suggested that injury likelihood is low when the acute:
chronic ratio is within a range of 0.8–1.3, and more than
double when the acute:chronic ratio exceeds 1.5 (Soligard
et al. 2016). Additional load measures such as high acute
weekly (Malone, Owen, et al. 2016), cumulated weekly
(Malone, Owen, et al. 2016;Rogalskietal.2013;Cross
et al. 2016) and week-to-week changes (Cross et al. 2016)
have also been associated with increased injury risk (IR),
whilst high chronic loads have been reported to offer pro-
tective effects (Hulin et al. 2016a). However, a concomitant
high acute:chonic ratio with high chronic load has been
shown to increase the risk (Hulin et al. 2016b).
Whilst the understanding of the relationships between
training load and IR in sports is increasing, some issues and
limitations of other studies should be addressed. For example,
studies on elite football players are lacking, with only one
study in male senior (Malone, Owen, et al. 2016) and another
in elite youth players (Bowen et al. 2017). It has also been
suggested that different acute:chronic durations should be
used to better fit the specificity of training/competitive pat-
terns of football (e.g., 2 and 3 weekly chronic loads) (Buchheit
2017). In addition, there is a lack of a standard injury definition
(Hulin 2017) with some studies investigating all injuries includ-
ing “medical attention injuries”(i.e., injuries without any time
loss) (Gabbett 2004), while others consider only time-loss
injuries (Malone, Owen, et al. 2016; Malone, Roe et al. 2016;
Hulin et al. 2016a,2016b). Moreover, despite earlier recom-
mendation (Gabbett 2010), only two studies have further clas-
sified time-loss injuries into contact and non-contact injury
mechanisms (Rogalski et al. 2013; Bowen et al. 2017). While
research has shown that non-contact injuries can be pre-
vented (to a certain extent) with exercise-based programmes
(Thorborg et al. 2017) in amateur football, there is still a lack of
CONTACT Maurizio Fanchini maurizio.fanchini@univr.it
SCIENCE AND MEDICINE IN FOOTBALL, 2018
https://doi.org/10.1080/24733938.2018.1429014
© 2018 Informa UK Limited, trading as Taylor & Francis Group
understanding on the relationship between load characteris-
tics and non-contact injuries. Therefore, there is a need to
further examine the relationships between load characteristics
and non-contact injuries specifically in elite football players.
One controversial aspect of previous research that requires
clarification is the supposition that significant associations
between load and injury have prompted load and its derivatives
to be deemed as “predictive”of injury (Hulin et al. 2016a,2016b).
However, as explained by Professor Roald Bahr (Bahr 2016),
associations with injury and predicting injury are different terms
and are analysis that should be used for different purposes (i.e.,
explanation and classification). Indeed, it has been shown that
even strong associations may not predict injury (Pepe 2004).
While associations such as odds ratio (OR) or relative risk can
be useful to characterize the risk of a population, these statistical
outcomes do not infer “prediction”(Pepe 2004). In order to
determine the true use of a marker, it is important to assess not
only an association but also the predictive validity and optimal
cut-off value (Fawcett 2006). Predictive validity of a marker is
typically assessed with receiver operating characteristic (ROC)
curve (De Vet et al. 2011).
Utilising a simple, practical and valid measure of internal
load derived from the rate of perceived exertion (RPE), the aim
of the present study was to investigate (1) the association of
internal perceived load with non-contact injury and (2) the
ability of these measures to predict injury in elite football
players. We hypothesised that there would be a significant
association between various load measures and non-contact
injury, though load measures would show inadequate level of
predictive validity.
Methods
Participants
During three consecutive, competitive seasons (2013/14, 2014/
15, 2015/16), 34 professional football players from the same
SerieA team were prospectively followed (age: 26 ± 5 y; height:
182 ± 5 cm; body mass: 78 ± 4 kg). Only players competing at
least 1 full season were included in the analysis, specifically, 12
players during three seasons, 10 during two seasons and 12 for
one season. Each season was split into three distinct periods
depending on the year and league start date: pre-season (early
July to mid-end August), early and late in-season (mid-end
August to end December and early January to mid-May, respec-
tively, early and late in-season). Pre-season period lasted
38 ± 3 days, the in-season (i.e., competitive) periods were
128 ± 4 and 147 ± 7 days in early and late in-season, respectively.
The study is in accordance with the spirit of the Helsinki
Declaration. Approval for the study was obtained from the Club
as player’s data were routinely collected over the course of the
season (Winter and Maughan 2009).
Quantification of internal load
Internal load was assessed using session-RPE method (Foster
et al. 2001) and validated in football (Impellizzeri et al. 2004)
whereby the intensity of the session is multiplied by the
duration for each player for each training session or match.
The intensity was determined using the Borg CR-10 scale
modified by Foster et al. (2001) within 30 min after completion
of the session/match and was collected by the same fitness
coach throughout all seasons. The mean weekly RPE was
calculated as indicator of the week intensity as well as the
total exposure was calculated by summating the duration of
each session. The weekly load (WL) corresponded to the sum
of load for all training sessions and matches for each week.
The absolute week-to-week change in load (W-WL) was calcu-
lated as the difference between current WL and previous WL.
Cumulative loads were calculated for 2, 3 and 4 weeks.
Chronic loads were calculated as rolling averages for 2, 3
and 4 weeks. An acute (WL) to chronic (rolling average of
previous weeks) ratio was calculated for 2, 3 and 4 weeks
(acute:chronic2, acute:chronic3, acute:chronic4, respectively).
Injury data collection
A non-contact injury was considered when a player was
unable to take a full part in future football training or match
due to physical complaints (Fuller et al. 2006) (all injuries
diagnosed by the same club doctor). Injuries were recorded
according to the classifications proposed by Fuller et al. (2006).
Training exposure was considered as the duration (h) corre-
sponding to team-based and individual physical activities.
Injury incidence was calculated as all non-contact injuries per
1000 h of football activity (training + matches).
Statistical analysis
Data are presented as mean ± standard deviations (mean ± SD).
Generally, in elite football teams, there are several differences in
the training structure between pre- and in-season periods,
including the organization, quality and quantity of training
(e.g., general vs. specific, once vs. twice a day training session)
and this results in different periodization and overall loads
(Impellizzeri et al. 2005). Differences in load between pre- and
in-season (early and late in-season, before and after Christmas
break, respectively) have been examined in the present cohort
with a one-way ANOVA. If WLs were different between periods
(P< 0.05), only in-season data were considered in the analysis.
Effect size (partial eta squared, η
2
) was also calculated and
values of 0.01, 0.06 and >0.15 were interpreted as small, med-
ium and large, respectively. Pearson’smomentofCorrelation
and variance inflation factors (VIF) were determined to detect
multicollinearity between markers; if a VIF ≥10 was found, the
variable was excluded from the analysis.
Determining association
A generalized estimating equations (GEE) analysis was used to
determine the association between load measures and non-
contact injury in the subsequent week (Williamson et al. 1996).
To analyse longitudinal data with a binary outcome distribu-
tion (injury: yes/no), a logit link function was used and
exchangeable working correlations matrix was chosen, based
on lower quasi-likelihood under the independence model cri-
terion. If GEE analysis was significant (P< 0.05), the variable
was split in four groups based on 15th, 50th, 85th percentile
to compare IR between zones of different load: extremely low
2M. FANCHINI ET AL.
(<15th percentile), moderate low (>15–50th percentile), mod-
erate high (from >50 to 85th percentile) and extremely high
(>85th percentile). The percentage IR was calculated as a
proportion of weeks with a subsequent injury for each group
(Hopkins et al. 2007).
ORs and 90% confidence intervals (CIs) were calculated for
comparison between risks in different load groups (MedCalc
Software, Ostend, Belgium). Magnitude-based inferences were
used to interpret change in percentage risk between groups
(Hopkins et al. 2009). The smallest beneficial and harmful effect
for a risk ratio was considered as an OR of <0.90 and >1.11,
respectively. The effect was considered unclear if the chance of
the true values of beneficial was >25% with the OR <66. If the effect
was considered clear, thresholds for assigning qualitative terms
of beneficial, trivial, harmful were as follows: <0.5%, most likely;
0.5–5%, very unlikely; 5–25%, unlikely; 25–75%, possible; 75–95%,
likely; 95–99.5%, very likely; >99.5%, most likely (Hopkins 2007).
Determining prediction
Load variables that exhibited a significant association follow-
ing GEE analysis were tested for predictive ability using ROC
curve. The ROC curve examines the discriminant ability of a
marker to classify players in two groups and plots the true
positive rate (sensitivity) against the true negative rate (speci-
ficity) producing an area under the curve (AUC). An AUC of
1.00 (100%) represents perfect discriminant power, where 0.50
(50%) would represent no discriminatory power (Fawcett
2006; Crowcroft et al. 2016). An AUC >0.70 and the lower CI
>0.50 was classified as a “good”benchmark (Menaspà et al.
2010). All ROC curve results were presented as AUC ± 90% CI.
To examine the predictive ability of the higher cut-off,
sensitivity [true positive/(true positive + false negative)*100]
and specificity [false positive/(false positive + true negative)
*100] were calculated for each load-based marker using the
85th percentile as cut-off (Kent and Hancock 2016). In addi-
tion, the Youden Index (J) was calculated (J= sensitivity+spe-
cificity−1) from all ROC curve plots to determine the point
where the sensitivity and specificity were optimised (i.e., high
J) and considered the score at which a “cut-off”value from
each load marker might be acceptable to discriminate a player
at risk of injury. The maximum Jindex of 1 would suggest
perfect discriminatory value, whilst a score of 0 would reflect
no diagnostic value (Schisterman et al. 2005).
All analyses were performed using SPSS (Version 21. IBM
Company, New York, USA) and Excel (Microsoft Excel 2011 for
Mac, Microsoft Corporation, USA).
Results
A total of 90 (of which 72 occurred in-season and 18 classified as
re-injuries) time-loss non-contact injuries were incurred during
the three seasons. Non-contact injury incidences were 5.2/
1000 h, 4.6/1000 h and 5.5/1000 h for pre-season, early in-season
and late in-season (5.0/1000 h for all in-season periods). Injury
occurrence, exposure and injury incidence per match and train-
ing are presented in Table 1. Details about type, location and re-
injuries are presented in Appendix 1. WL was significantly differ-
ent at different time points during the season (P<0.0001,
η
2
=0.14).Pre-seasonshowedhighWLcomparedtoearlyand
late in-season periods (2082 ± 700 vs. 1528 ± 466 AU, P< 0.0001
and vs. 1526 ± 427 AU, P< 0.0001, respectively). Pre-season data
were excluded from the analysis; therefore, 72 non-contact inju-
ries were considered. Cumulative load of 2, 3 and 4 weeks
showed substantial multi-collinearity (i.e., VIF >10) and therefore
were excluded from the analysis.
Association
The results of the GEE analysis showed (Table 2)thatweekload
and week-to-week load absolute change were significantly but
Table 1. Injury occurrence, exposure and injury incidence during all seasons (pre- and in-season) per training and match.
2013/14 2014/15 2015/16
Pre-season In-season Pre-season In-season Pre-season In-season
Non-contact injuries, training (n°) 7 17 6 7 5 15
Non-contact injuries, match (n°) 9 14 10
Exposure, training (h) 840 4146 1309 4309 1295 4134
Exposure, match (h) 494 592 593
Injury incidence, training (n°/1000 h) 8.3 4.1 4.6 1.6 3.9 3.6
Injury incidence, match (n°/1000 h) 18.2 23.7 16.9
Table 2. Association and prediction of different load markers. Odds ratios, 90% confidence intervals (90% CI) and P-level from the Generalized Estimating Equation
analysis for load-based markers and binary outcome (injury: yes/no). Area Under the Curve (AUC) and 90% CI from the Receiving Operator Characteristic curve (ROC)
analysis and the Youden Index (J).
Association Prediction
Marker OR (90% CI) P-level Clinical inference AUC (90% CI) J
Duration (min) 1.00 (1.00–1.00) 0.08 Unclear
RPE (AU) 1.37 (0.97–1.93) 0.13 Likely harmful 0.54 (0.48–0.59) 0.11
WL (AU) 1.00 (1.00–1.00) 0.02 Unclear 0.55 (0.50–0.59) 0.15
W-WL (AU) 1.00 (1.00–1.00) 0.04 Unclear 0.56 (0.51–0.62) 0.14
Acute:chronic2 (AU) 2.98 (1.87–4.75) 0.00 Most likely harmful 0.57 (0.52–0.63) 0.15
Acute:chronic3 (AU) 2.46 (1.43–4.24) 0.01 Very likely harmful 0.60 (0.54–0.65) 0.19
Acute:chronic4 (AU) 2.91 (1.58–5.36) 0.00 Most likely harmful 0.57 (0.52–0.63) 0.15
RPE: rate of perceived exertion; WL: week load; W-WL: week-to-week absolute difference in load; Acute:chronic2, Acute:chronic3, Acute:chronic4 WL ratios; OR: odds
ratio; AUC: Area Under the Curve; J: Youden Index.
SCIENCE AND MEDICINE IN FOOTBALL 3
unclearly associated with injuries whereas acute:chronic2, acute:
chronic3, acute:chronic4 showed clear association with injuries
(P< 0.05). Weekly duration and RPE showed no significant
(unclear and clear, respectively) associations with injuries
(Table 2). The IR, OR and 90% CI calculated for comparison
between different load groups in each load marker (<15th,
15–50th, 50–85th, >85th percentile) and corresponding magni-
tude-based inferences are presented in Table 3.
Prediction
The ROC curve (Figure 1), the values AUC (90% CI) and the J
for each load marker (Table 2) showed poor predictive ability
of injury (AUC: 0.55–0.60). When using the 85th percentile as
cut-off, each marker showed poor sensitivity (Table 4). Given
the decrease of IR in the higher group of the WL compared to
intermediate groups (i.e., 3.1% vs. 4.5%, 4.1%, respectively in
4th, 2th, 3th group), a further analysis was performed to
examine the high frequency of non-contact injuries group
Table 3. Injury risk comparisons between different zones of load (<15th, 15–50th, 50–85th, >85th percentile).
Marker (AU)
Injury risk
(%) Load zones
Odds ratio
(90% CI)
Qualitative term for clinical
inference
Chances (%) the effect is
beneficial/trivial/harmful
WL 1.4 <1086 (reference)
4.5 1086 to 1542 3.4 (1.42–8.28) Very likely Harmful 1/1/98
4.1 >1542 to 1985 3.1 (1.27–7.50) Very likely Harmful 1/2/97
3.1 >1985 2.3 (0.84–6.19) Likely harmful 6/6/88
1086 to 1542 (reference)
>1542 to 1985 0.9 (0.58–1.40) Unclear 50/29/22
>1985 0.7 (0.35–1.25) Unclear 78/12/9
>1542 to 1985 (reference)
>1985 0.7 (0.39–1.40) Unclear 69/16/15
W-WL 2.7 <–572 (reference)
3.4 −572 to 1 1.2 (0.63–2.47) Possibly harmful 22/17/61
4.1 >1 to 614 1.5 (0.78–2.97) Likely harmful 10/12/78
4.5 >614 1.7 (0.78–3.51) Likely harmful 9/10/81
−572 to 1 (reference)
>1 to 614 1.2 (0.76–1.96) Possibly harmful 14/23/63
>614 1.3 (0.74–2.38) Possibly harmful 14/17/69
>1 to 614 (reference)
>614 1.1 (0.62–1.91) Unclear 29/23/47
Acute:chronic2 2.4 <0.81 (reference)
3.5 0.81 to 1.00 1.5 (0.73–3.05) Possibly harmful 12/13/75
3.8 >1.00 to 1.20 1.6 (0.79–3.29 Likely harmful 9/11/81
5.1 >1.20 2.2 (1.03–4.74) Likely harmful 3/4/93
0.81 to 1.00 (reference)
>1.00 to 1.20 1.1 (0.68–1.74) Unclear 26/27/47
>1.20 1.5 (0.85–2.58) Likely harmful 7/13/80
>1.00 to 1.20 (reference)
>1.20 1.4 (0.79–2.36) Possibly harmful 11/16/73
Acute:chronic3 2.4 <0.80 (reference)
2.7 0.80 to1.01 1.1 (0.53–2.34) Unclear 32/18/50
4.4 >1.01 to 1.23 1.9 (0.94–3.81) Likely harmful 4/7/89
5.9 >1.23 2.5 (1.20–5.38) Very likely harmful 1/2/96
0.80 to1.01 (reference)
>1.01 to 1.23 1.7 (1.03–2.79) Likely harmful 2/6/92
>1.23 2.3 (1.29–4.01) Very likely harmful 0/2/98
>1.01 to 1.23 (reference)
>1.23 1.3 (0.80–2.24) Possibly harmful 10/17/73
Acute:chronic4 1.4 <0.78 (reference)
3.3 0.78–1.02 2.4 (0.98–5.92) Likely harmful 4/4/92
4.5 >1.02–1.26 3.3 (1.37–8.02) Very likely harmful 1/1/98
4.9 >1.26 3.6 (1.41–9.31) Very likely harmful 1/1/98
0.78–1.02 (reference)
>1.02–1.26 1.4 (0.86–2.21) Likely harmful 7/16/78
>1.26 1.5 (0.85–2.68) Likely harmful 7/12/81
>1.02–1.26 (reference)
>1.26 1.1 (0.63–1.89) Unclear 28/24/48
WL: week load; W-WL: week-to-week absolute change in week load; Acute:chronic2, Acute:chronic3, Acute:chronic4 WLs ratio.
1 - Specificity
1.00.80.60.40.20.0
Sensitivity
1.0
0.8
0.6
0.4
0.2
0.0
Reference Line
Acute:chronic2
Acute:chronic3
Acute:chronic4
Source of the
Curve
Figure 1. Receiving operating characteristic (ROC) curves for the acute:chronic2,
acute:chronic3, acute:chronic4 WL ratios markers.
4M. FANCHINI ET AL.
compared to the other (i.e., WL between >1086 and 1542 vs.
other groups), and sensitivity and specificity were 43.1% and
65.3%, respectively.
Discussion
In accordance with our hypotheses, the main and novel find-
ings of this study were that various load markers were clearly
associated with the occurrence of non-contact injury in elite
footballers; however, they showed poor predictive ability to
detect individuals that will go on to incur a non-contact injury.
Evidence for implementing a global load monitoring
strategy
The results of the present study support the IOC consensus
statement (Soligard et al. 2016), which highlights the asso-
ciation of load and IR. However, despite being significantly
associated (P< 0.05) with non-contact injury, the absolute
loads in the current study (i.e., weekly acute load and week-
to-week absolute change) did not exhibit an increased risk
of injury (OR 1, unclear clinical inference). Despite RPE (i.e.,
intensity) showing no significant association (P> 0.05) with
non-contact injury using traditional statistical analyses, mag-
nitude-based inference statistics highlighted a likely harmful
effect (Table 2). As recommended by the IOC, it may be the
rate of load application that is a more pertinent measure
associated with injury. The acute:chronic ratios of 2, 3 and
4 weeks investigated in the current study all showed a
significant association with non-contact injury coupled with
an increased risk (OR 3.0, 2.5 and 2.9, respectively). These
results support the findings in other sports that the acute:
chronic ratio offers some explanation into non-contact injury
occurrence with rapid increases in relative load (Soligard
et al. 2016). Unfortunately, our results showed similar asso-
ciation with IR in acute:chronic 2, 3, 4 ratios; therefore,
further work is needed not only to validate our findings
but also to identify the optimal combination of acute and
chronic durations. Recently, a protective effect of a tradi-
tional acute:chronic4 between 1.00 and 1.25 has been found
in elite football players (Malone, Owen, et al. 2016).
However, the present study investigated OR between differ-
ent load groups (i.e., based on percentiles) for all acute:
chronic ratios and it was found that IR increased (OR >1)
as the acute:chronic values increased (Table 3). Collectively,
these results show that there was no protective effect of
load in the present cohort.
But, can we predict when a player will get injured?
While the findings of significant association with non-contact
injury provides the practitioner with evidence to implement a
load monitoring preventative strategy, such information does
not imply an ability to predict individual players who will incur
the injury (i.e., diagnostic accuracy). The poor diagnostic char-
acteristics of load variables in the present study (Table 2) show
that load monitoring cannot be confidently used as a tool to
predict injury in individual players. In the present sample, ROC
curve analysis revealed AUC ≤0.60 for WL, week-to-week
change as well as acute:chronic2, 3 and 4 (Table 2), which is
lower than the AUC >0.70 which has to be reported to estab-
lish some predictive ability (Crowcroft et al. 2016). In addition,
the Youden Index, which assesses the balance between sensi-
tivity and specificity to quantify cut-off values (Crowcroft et al.
2016), also showed poor discriminatory value (J= 0.14–0.19,
Table 2). A Jof 1 suggests a perfect discriminatory value and J
of 0 reflects no diagnostic value (Schisterman et al. 2005);
unfortunately, our results were close to 0. While, to our knowl-
edge (Schisterman et al. 2005), this is the first study to use a
predictive analysis for load variables as predictors of injury
occurrence in elite athletes, it is in line with other studies
showing poor diagnostic ability of subjective monitoring vari-
ables used to identify performance change (Crowcroft et al.
2016; Saw et al. 2016). In addition, mathematical coupling has
been found between numerator and denominator in the
acute:chronic ratio (i.e., acute is a part of the chronic load)
providing spurious correlations and therefore limiting infer-
ences (Lolli et al. 2017).
What if the acute:chronic ratio is really high?
In the present study, we investigated the sensitivity (i.e., the
ability of this load measure to correctly identify a player
incurring non-contact injury) of very high acute:chronic ratios
(>85th percentile versus ≤85th percentile). With the acute:
chronic2, 3 and 4, we found 21%, 24% and 20% sensitivity,
respectively. To put this into perspective, in absolute number
of non-contact injuries, acute:chronic2 identified 15 true posi-
tives and 277 false positives, acute:chronic3: 17 true positives
and 273 false positives and acute:chronic4: 14 true positives
and 274 false positives (Table 4). Given these findings, even
with very high acute:chronic ratios (considering the present
sample), it seems unlikely to predict injury. It is important to
highlight that the value of acute:chronic4 > 85th percentile in
the present study is lower compared to that reported in the
Table 4. Sensitivity and specificity for each marker when the load is extremely high (i.e., >85th percentile).
Variables
(AU) True positive False positive False negative True negative Sensitivity (%) Specificity (%)
WL 9 284 63 1597 12.5 84.9
WL 2nd vs. other 31 653 41 1229 43.1 65.3
W-WTL 13 279 59 1593 18.1 85.1
Acute:chronic2 15 277 57 1596 20.8 85.2
Acute:chronic3 17 273 55 1588 23.6 85.3
Acute:chronic4 14 274 56 1576 20.0 85.2
True positive: the players with injury with marker >85th percentile; False positive: the players without injury with marker >85th percentile; False negative: the
players with injury with marker <85th percentile; True negative: the players without injury with marker <85th percentile. Sensitivity and Specificity.
SCIENCE AND MEDICINE IN FOOTBALL 5
IOC consensus (1.26 vs. 1.50 AU, respectively) and this may
explain the difference in results (Soligard et al. 2016).
Limitations and future directions
There are some limitations to the current study. First, the
cohort used in this study belong to one team and limits
generalization of the findings. Future studies should include
more teams over multiple seasons. Second, we did not
account for potential confounding variables such as previous
injury, multiple injuries in the same player, fitness or age.
Future studies would benefit from methods and analyses
that account for these confounders. In addition, the present
study examined only internal load while other studies exam-
ined external loads (i.e., running loads); future studies should
examine both internal and external measures together. The
strengths of this study include a longitudinal period studied
over three seasons and statistical analyses accounting for
repeated measures in addition to investigating predictive
ability.
Conclusion
The findings of the present study support the association
between session-RPE derived training and match load mea-
sures, especially relative measures of acute:chronic2, 3 and 4
with non-contact injury. However, while significantly asso-
ciated, this should not be confused with ability to predict
injury at an individual player level. Overall, our findings pro-
vide justification for the implementation of a team-wide mon-
itoring strategy of internal load in elite footballers; however,
caution should be taken when making decisions at the indivi-
dual player level.
Practical implications
The load markers based on internal perceived load (session-
RPE) as acute:chronic ratios (e.g., 1:4 weeks, 1:3 weeks and 1:2
weeks) are significantly associated with non-contact injury in
elite football players. While internal perceived load markers
are associated with non-contact injury occurrence, such mar-
kers have poor predictive validity to identify individual players
who will go on to incur such an injury.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Maurizio Fanchini http://orcid.org/0000-0001-7398-5443
Aaron J. Coutts http://orcid.org/0000-0002-1782-7691
References
Bahr R. 2016. Why screening tests to predict injury do not work—and
probably never will. ..: a critical review. Br J Sports Med. 50(13):776–780.
Bowen L, Gross AS, Gimpel M, Li F-X. 2017. Accumulated workloads and
the acute: chronicworkload ratio relate to injury risk in elite youth
football players. Br J Sports Med. 51(5):452–459.
Buchheit M. 2017. Applying the acute: chronicworkload ratio in elite foot-
ball: worth the effort? Br J Sports Med. 51(18):1325–1327.
Cross MJ, Williams S, Trewartha G, Kemp SPT, Stokes KA. 2016. The
influence of in-season training loads on injury risk in professional
Rugby Union. Int J Sports Physiol Perform. 11(3):350–355.
Crowcroft S, McCleave E, Slattery K, Coutts AJ. 2016. Assessing the mea-
surement sensitivity and diagnostic characteristics of athlete monitor-
ing tools in national swimmers. Int J Sports Physiol Perform. 12(Suppl
2):1–21.
De Vet H, Terwee CB, Mokkink L, Knol D. 2011. Measurement in medicine:
a practical guide (Practical Guides to Biostatistics and Epidemiology).
Cambridge: Cambridge University Press; p. 167-168. doi:10.1017/
CBO9780511996214.
Fawcett T. 2006. An introduction to ROC analysis. Pattern Recognit Lett. 27
(8):861–874.
Foster C, Florhaug JA, Franklin J, Gottschall L, Hrovatin LA, Parker S,
Doleshal P, Dodge C. 2001. A new approach to monitoring exercise
training. J Strength Cond Res. 15(1):109–115.
Fuller CW, Ekstrand J, Junge A, Andersen TE, Bahr R, Dvorak J, Hagglund M,
McCrory P, Meeuwisse WH. 2006. Consensus statement on injury defi-
nitions and data collection procedures in studies of football (soccer)
injuries. Br J Sports Med. 40(3):193–201.
Gabbett TJ. 2004. Reductions in pre-season training loads reduce training
injury rates in rugby league players. Br J Sports Med. 38(6):743–749.
Gabbett TJ. 2010. The development and application of an injury prediction
model for noncontact, soft-tissue injuries in elite collision sport ath-
letes. J Strength Cond Res. 24(10):2593–2603.
Hopkins WG. 2007. A spreadsheet for deriving a confidence interval,
mechanistic inference and clinical inference from a P value.
Sportscience. 11:16–20.
Hopkins WG, Marshall SW, Batterham AM, Hanin J. 2009. Progressive
statistics for studies in sports medicine and exercise science. Med Sci
Sports Exerc. 41(1):3–13.
Hopkins WG, Marshall SW, Quarrie KL, Hume PA. 2007. Risk factors and risk
statistics for sports injuries. Clin J Sport Med. 17(3):208–210.
Hulin BT. 2017. The never-ending search for the perfect acute: chronicwork-
load ratio: what role injury definition? Br J Sports Med. 51(13):991–992.
Hulin BT, Gabbett TJ, Blanch P, Chapman P, Bailey D, Orchard JW. 2014.
Spikes in acute workload are associated with increased injury risk in
elite cricket fast bowlers. Br J Sports Med. 48(8):708–712.
Hulin BT, Gabbett TJ, Caputi P, Lawson DW, Sampson JA. 2016a. Low
chronic workload and the acute: chronicworkload ratio are more pre-
dictive of injury than between-match recovery time: a two-season
prospective cohort study in elite rugby league players. Br J Sports
Med. 50(16):1008–1012.
Hulin BT, Gabbett TJ, Lawson DW, Caputi P, Sampson JA. 2016b. The acute:
chronicworkload ratio predicts injury: high chronic workload may
decrease injury risk in elite rugby league players. Br J Sports Med. 50
(4):231–236.
Impellizzeri FM, Rampinini E, Coutts AJ, Sassi A, Marcora SM. 2004. Use of RPE-
based training load in soccer. Med Sci Sports Exerc. 36(6):1042–1047.
Impellizzeri FM, Rampinini E, Marcora SM. 2005. Physiological assessment
of aerobic training in soccer. J Sports Sci. 23(6):583–592.
Kent P, Hancock MJ. 2016. Interpretation of dichotomous outcomes: sen-
sitivity, specificity, likelihood ratios, and pre-test and post-test prob-
ability. J Physiother. 62(4):231–233.
Lolli L, Batterham AM, Hawkins R, Kelly DM, Strudwick AJ, Thorpe R,
Gregson W, Atkinson G. 2017. Mathematical coupling causes spurious
correlation within the conventional acute-to-chronic workload ratio
calculations. Br J Sports Med.
Malone S, Owen A, Newton M, Mendes B, Collins KD, Gabbett TJ. 2016. The
acute: chonicworkload ratio in relation to injury risk in professional
soccer. J Sci Med Sport. 20(6):561–565.
MaloneS,RoeM,DoranDA,GabbettTJ,CollinsK.2016. High chronic
training loads and exposure to bouts of maximal velocity running
reduce injury risk in elite Gaelic football. J Sci Med Sport. 20
(3):250–254.
Menaspà P, Sassi A, Impellizzeri FM. 2010. Aerobic fitness variables do not
predict the professional career of young cyclists. Med Sci Sports Exerc.
42(4):805–812.
6M. FANCHINI ET AL.
Murray NB, Gabbett TJ, Townshend AD, Hulin BT, McLellan CP. 2016.
Individual and combined effects of acute and chronic running loads
on injury risk in elite Australian footballers. Scand J Med Sci Sports. 27
(9):990–998.
Pepe MS. 2004. Limitations of the odds ratio in gauging the performance
of a diagnostic, prognostic, or screening marker. Am J Epidemiol. 159
(9):882–890.
Rogalski B, Dawson B, Heasman J, Gabbett TJ. 2013. Training and game
loads and injury risk in elite Australian footballers. J Sci Med Sport. 16
(6):499–503.
Saw AE, Main LC, Gastin PB. 2016. Monitoring the athlete training
response: subjective self-reported measures trump commonly used
objective measures: a systematic review. Br J Sports Med. 50(5):281–
291.
Schisterman EF, Perkins NJ, Liu A, Bondell H. 2005. Optimal cut-point and
its corresponding Youden Index to discriminate individuals using
pooled blood samples. Epidemiology. 16(1):73–81.
Soligard T, Schwellnus M, Alonso J-M, Bahr R, Clarsen B, Dijkstra HP, Gabbett
T, Gleeson M, Hägglund M, Hutchinson MR, et al. 2016. How much is too
much? (Part 1) International Olympic Committee consensus statement on
load in sport and risk of injury. Br J Sports Med. 50(17):1030–1041.
Thorborg K, Krommes KK, Esteve E, Clausen MB, Bartels EM, Rathleff MS.
2017. Effect of specific exercise-based football injury prevention pro-
grammes on the overall injury rate in football: a systematic review and
meta-analysis of the FIFA 11 and 11+ programmes. Br J Sports Med. 51
(7):562–571. eng.
Weiss KJ, Allen SV, McGuigan MR, Whatman CS. 2017.Therelationship
between training load and injury in men’s professional Basketball players.
Int J Sports Physiol Perform. 12(9):1238-1242. doi:10.1123/ijspp.2016-0726.
Williamson DS, Bangdiwala SI, Marshall SW, Waller AE. 1996. Repeated
measures analysis of binary outcomes: applications to injury research.
Accid Anal Prev. 28(5):571–579.
Winter EM, Maughan RJ. 2009. Requirements for ethics approvals. J Sports
Sci. 27(10):985.
Appendix 1. Details of re-injuries during pre- and in-season periods*
Season Location Type of injury Event (Training/match)
2013/14
Pre-season Thigh (H) Muscle strain Training
Thigh (Q) Muscle strain Training
Knee Sprain Training (Friendly match)
Thigh (Ad) Muscle strain Training (Friendly match)
Thigh (H) Muscle strain Training (Friendly match)
Thigh (H) Muscle strain Training (Friendly match)
In-season Knee Lesion of meniscus Training
Hip Muscle tear Training
2014/15
Pre-season Lower back Muscle strain Training
Lower leg Muscle strain Training
Lower leg Muscle strain Training
Thigh (H) Muscle strain Training (Friendly match)
Thigh (Ad) Muscle strain Training
Lower leg Muscle strain Training
In-season Knee Ligament injury Match
Thigh (H) Muscle tear Match
Thigh Muscle strain Match
Thigh (Q) Muscle strain Match
2015/16
Pre-season Thigh (Q) Muscle tear Training (Friendly match)
Knee (PT) Tendinosis Training
Lower leg Muscle strain Training
Foot Other injuries Training
Thigh (H) Muscle strain Training (Friendly match)
In-season Knee (CL) Ligament injury Training
Thigh (H) Muscle strain Match
Lower leg/Achilles tendon Tendinosis Training
Thigh (H) Muscle strain Match
Thigh (H) Muscle tear Match
Knee (PT) Tendinosis Training
Knee (PT) Tendinosis Training
Groin Muscle strain Training
Thigh Muscle strain Training
Thigh (H) Muscle tear Match
Thigh (H) Muscle tear Match
Knee Other bone injuries Match
H: hamstring; Q: quadriceps; Ad: adductor; PT: patellar tendon; CL: collateral ligament.
*Fuller CW, Ekstrand J, Junge A, Andersen TE, Bahr R, Dvorak J, Hagglund M, McCrory P, Meeuwisse WH. 2006. Consensus statement
on injury definitions and data collection procedures in studies of football (soccer) injuries. Br J Sports Med. 40(3):193–201.
SCIENCE AND MEDICINE IN FOOTBALL 7
