Content uploaded by Tim Gabbett
Author content
All content in this area was uploaded by Tim Gabbett on Jul 05, 2020
Content may be subject to copyright.
THE DEVELOPMENT AND APPLICATION OF AN INJURY
PREDICTION MODEL FOR NONCONTACT,SOFT-TISSUE
INJURIES IN ELITE COLLISION SPORT ATHLETES
TIM J. GABBETT
1,2
1
Brisbane Broncos Rugby League Club, Brisbane, Australia; and
2
School of Human Movement Studies,
The University of Queensland, Brisbane, Australia
ABSTRACT
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 24(10): 2593–2603,
2010—Limited information exists on the training dose–response
relationship in elite collision sport athletes. In addition, no study
has developed an injury prediction model for collision sport
athletes. The purpose of this study was to develop an injury
prediction model for noncontact, soft-tissue injuries in elite
collision sport athletes. Ninety-one professional rugby league
players participated in this 4-year prospective study. This study
was conducted in 2 phases. Firstly, training load and injury data
were prospectively recorded over 2 competitive seasons in elite
collision sport athletes. Training load and injury data were
modeled using a logistic regression model with a binomial
distribution (injury vs. no injury) and logit link function. Secondly,
training load and injury data were prospectively recorded over
a further 2 competitive seasons in the same cohort of elite
collision sport athletes. An injury prediction model based on
planned and actual training loads was developed and
implemented to determine if noncontact, soft-tissue injuries
could be predicted and therefore prevented in elite collision
sport athletes. Players were 50–80% likely to sustain a pre-
season injury within the training load range of 3,000–5,000
units. These training load ÔthresholdsÕwere considerably
reduced (1,700–3,000 units) in the late-competition phase of
the season. A total of 159 noncontact, soft-tissue injuries were
sustained over the latter 2 seasons. The percentage of true
positive predictions was 62.3% (n= 121), whereas the total
number of false positive and false negative predictions was 20
and 18, respectively. Players that exceeded the training load
threshold were 70 times more likely to test positive for
noncontact, soft-tissue injury, whereas players that did not
exceed the training load threshold were injured 1/10 as often.
These findings provide information on the training dose–
response relationship and a scientific method of monitoring and
regulating training load in elite collision sport athletes.
KEY WORDS injury prevention, rugby league, training monitoring,
training load, applied sport science
INTRODUCTION
The training–performance relationship is of partic-
ular importance to coaches to determine the
optimum amount of training required to attain
specific performance levels (2,5,11). Bannister et al.
(4) proposed a statistical model to describe an athlete’s
response to a given training stimulus. According to this
model, the performance of an athlete in response to training
can be estimated from the difference between a negative
function (fatigue) and a positive function (fitness). Studies
have described the training–performance relationship as
analogous with the dose–response relationship reported in
pharmacological studies, with the primary goal of providing
a training stimulus that maximizes performance potential and
minimizes the negative consequences of training (i.e., injury,
illness, fatigue, overtraining) (29).
Several studies have investigated the influence of training
volume, intensity, and frequency on athletic performance,
with performance generally improving with increases in
training load (11,12,28,31,32,37,39). Studies of the training–
performance relationship in individual sports (e.g., swimming
and running) have found a positive relationship between both
greater training volume and performance (12,39) and higher
training intensity and performance (28,32,37). Foster et al.
(11) studied 56 runners, cyclists, and speed skaters during 12
weeks of training and reported that a 10-fold increase in
training load was associated with a ;10% improvement in
performance. Moreover, Stewart and Hopkins (39)
reported a significant relationship between greater training
volume and performance (r= 0.50–0.80) and higher training
intensity and performance (r= 0.60–0.70) in competitive
swimmers. However, it has also been shown that negative
adaptations to exercise training are dose related, with the
Address correspondence to Dr. Tim J. Gabbett, timg@broncos.com.au.
24(10)/2593–2603
Journal of Strength and Conditioning Research
Ó2010 National Strength and Conditioning Association
VOLUME 24 | NUMBER 10 | OCTOBER 2010 | 2593
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
highest incidence of illness and injury occurring when
training loads are highest (10,17).
In contrast to individual sports, collision sports (e.g., ice
hockey, rugby, and lacrosse) are characterized by large
numbers of physical collisions and tackles, short repeated
sprints, rapid acceleration, deceleration, and changes of
direction, and an ability to produce high levels of muscular
force extremely rapidly (19). As a result, collision sport
athletes are required to have well-developed speed, strength,
muscular power, agility, and maximal aerobic power
(
_
VO
2
max). Previous studies of collision sport athletes have
reported a significant relationship (r= 0.86) between training
loads and training-injury rates (17), suggesting that the harder
these athletes train, the more injuries they will sustain.
Furthermore, reductions in training loads have been shown
to reduce training-injury rates and result in greater improve-
ments in
_
VO
2
max (18). However, it has also been shown that
collision sport athletes that perform ,18 weeks of preseason
training before sustaining an initial injury are at an increased
risk of sustaining a subsequent injury, whereas players with
a low off-season
_
VO
2
max are at an increased risk of sustaining
a contact injury (20). Clearly, training for collision sports
reflects a balance between the minimum training load
required to elicit an improvement in fitness and the
maximum training load tolerable before sustaining marked
increases in injury rates.
The majority (;60%) of collision sport injuries have been
shown to occur in physical collisions and tackles (14,15,23,24)
and are therefore thought to be mostly unavoidable.
However, a considerable proportion of injuries sustained
by collision sport athletes are noncontact, soft-tissue issues
that occur as a result of excessive training loads, inadequate
recovery, and overtraining (15,17–19,21). These injuries,
which are largely preventable, have the potential to impact
team selections and as a result may influence team perfor-
mance. Therefore, an obvious challenge for coaches is to
identify the athletes that have received an adequate training
stimulus to compete optimally and those that may be
susceptible to overreaching or overtraining. However, no
study has identified the optimum training load to minimize
noncontact, soft-tissue injuries in collision sport athletes.
Although models of the training–performance relationship
have been constructed for athletes from individual sports
(2–4,10,11,30,31,39), studies of the training–performance
relationship of collision sport athletes are limited (9,26)
and have been confined to the identification of hormonal and
psychological markers of overtraining and fatigue. In
addition, studies examining the influence of training load
on injury in collision sports have been limited to subelite
athletes (21). In the elite team-sport environment, it is critical
to have the maximum number of players free from injury and
available for selection in as many games as possible through-
out the season (34,36). Injuries that result in lost playing time
may alter the team structure, leading to reduced cohesion
between players, and subsequent reductions in playing
performance (16). However, to date, limited information
exists on the training dose–response relationship in elite
collision sport athletes. In addition, no study has developed
an injury prediction model for collision sport athletes. With
this in mind, the purpose of this study was to develop an
injury prediction model for noncontact, soft-tissue injuries in
elite collision sport athletes.
METHODS
Experimental Approach to the Problem
This study involved (a) the collection of training load and
noncontact, soft-tissue injury data (over 2 competitive
seasons); (b) the modeling of training load with injury, to
determine the relationship between these 2 variables (i.e.,
with a given training load, what is the risk of noncontact, soft-
tissue injury?); and (c) the development and application of an
injury prediction model encompassing planned and actual
training loads (over a subsequent 2 seasons). The study was
conducted in 2 phases: Firstly, training load and injury data
were prospectively recorded over 2 competitive seasons in
elite collision sport athletes. Training load and injury data
were modeled using a logistic regression model with
a binomial distribution (injury vs. no injury) and logit link
function. This model was identical to that previously used to
investigate the training–injury relationship in subelite colli-
sion sport athletes (21). The development of this model
provided statistical information on the likelihood of soft-
tissue injury with a given training load, throughout the dif-
ferent phases of the season (i.e., preseason, early competition,
late competition). Secondly, training load and injury data
were prospectively recorded over a further 2 competitive
seasons in the same cohort of elite collision sport athletes.
Based on the results of the logistic regression model, an injury
prediction model encompassing planned and actual training
loads was developed and implemented to determine if non-
contact, soft-tissue injuries could be predicted and therefore
prevented in elite collision sport athletes. The proportion of
true positive (i.e., predicted injury and player sustained injury)
and negative (i.e., no injury predicted and the player did not
sustain injury) results, and false positive (i.e., predicted injury
but the player did not sustain any injury) and negative (i.e., no
injury predicted but the player sustained an injury) results
were also calculated to describe errors made in the statistical
decision process and to allow the calculation of sensitivity
(i.e., the proportion of injured players who were predicted to
be injured) and specificity (i.e., the proportion of uninjured
players who were predicted to remain injury-free) likelihood
ratios (1).
Subjects
Ninety-one professional rugby league players (mean 6SD
age, height, and body mass; 23.7 63.8 years, 183.2 64.9 cm,
and 94.4 69.2 kg, respectively) participated in this 4-year
prospective study (2006–2009). Of the 91 players, 53 (58.2%)
played 1 season, 22 players (24.2%) played 2 seasons,
2594
Journal of Strength and Conditioning Research
the
TM
Injury Prediction in Collision Sport Athletes
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
7 players (7.7%) played 3 seasons, whereas 9 players (9.9%)
played all 4 seasons. The number of players participating in
each season was 36, 38, 35, and 45 respectively, giving a total
of 154 player seasons. All players were highly motivated
players from the same professional rugby league club and
were competing in the elite National Rugby League
competition. Players had completed a 4-week active recovery
off-season period and were free from injury at the
commencement of the study. All players received a clear
explanation of the study, including the risks and benefits of
participation, and written consent was obtained. The
Institutional Review Board for Human Investigation ap-
proved all experimental procedures.
Data Collection
This study was conducted in 2 phases: Firstly, training load
and injury data were prospectively recorded over 2 compet-
itive seasons (2006 and 2007). Training load and injury data
were modeled to determine the relationship between training
load and the likelihood of injury (21). Secondly, training load
and injury data were prospectively recorded over a further 2
competitive seasons (2008 and 2009). During this period, an
injury prediction model based on planned and actual training
loads was developed and implemented to determine if
noncontact, soft-tissue injuries could be predicted and
therefore prevented in elite collision sport athletes.
Training Sessions
A periodized, game-specific training program was imple-
mented, with training loads progressively increased in the
general preparatory phase of the season (i.e., November to
February) and reduced during the competitive phase of the
season (i.e., March to October). The training program
progressed from high volume–low intensity activities during
the preseason conditioning period to low volume–high
intensity activities during the in-season conditioning period.
Each player participated in up to 5 organized field-training
sessions, and 4 strength sessions per week in the preseason
period, and 2–4 field-training sessions, and 1–2 strength
sessions per week in the competitive phase of the season.
Training load and injury data were recorded for every session.
To ensure training specificity, players were allocated into 1
of 3 training groups (i.e., hit-up forwards, adjustables, and
outside backs) according to the skills and physiological
demands of their individual positions. The content of sessions
was based largely on perceived strengths and weaknesses
within the club. The training sessions consisted of specific
skills, speed, muscular power, agility, and endurance training
common to rugby league. Skills sessions were designed to
develop passing and catching skills, defensive line speed and
technique, support play, and ball control. Although some
differences existed in the intensity of activities performed
throughout the season, the types of activities performed in the
preseason training phase (e.g., basic skills, light and full
contact tackling drills and longer interval running) were
similar to the early-competition and late-competition training
phases (e.g., light contact tackling drills, advanced skills, and
shorter repeated-sprint training). The duration of training
sessions was recorded, with sessions typically lasting between
60 and 120 minutes. An outline of the training plan for the
season is shown in Table 1.
Quantification of Training Loads
The intensity of individual training sessions was estimated
using a modified rating of perceived exertion (RPE) scale (13).
Training load was calculated by multiplying the training
session intensity by the duration of the training session and
was reported in arbitrary units. Intensity estimates were
obtained 30 minutes after completing the training session.
When compared to heart rate and blood-lactate concentra-
tion, the RPE scale has been shown to provide a valid
estimate of exercise intensity (8,10,27). In addition, before
commencing the study, we investigated the relationship
between heart rate and RPE, and blood-lactate concentration
and RPE on a subset of subjects during typical rugby league
training activities. The correlations between training heart
rate and training RPE, and training blood-lactate
TABLE 1. Outline of yearly training plan for elite collision sport athletes.*
Training phase Objective
Preseason Develop speed, agility, and muscular power
Develop aerobic endurance, anaerobic capacity, and repeated-effort ability
Develop game-specific individual and team skills
Early-competition Continue to develop game-specific skills
Maintain aerobic endurance, anaerobic capacity, repeated-effort ability,
speed, agility, and muscular power
Late-competition Maintain game-specific skills
Maintain aerobic endurance, anaerobic capacity, repeated-effort ability,
speed, agility, and muscular power
*Preseason = 16 weeks; early-competition = 15 weeks; late-competition = 15 weeks.
VOLUME 24 | NUMBER 10 | OCTOBER 2010 | 2595
Journal of Strength and Conditioning Research
the
TM
|
www.nsca-jscr.org
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
concentration and training RPE, were 0.89 and 0.86,
respectively (21). A subset of players (n= 11) also completed
2 identical off-season training sessions, performed 1 week
apart, before the commencement of the study, to determine
test–retest reliability. The intraclass correlation coefficients
for test–retest reliability and typical error of measurement for
the RPE scale were 0.99 and 4.0%, respectively. Collectively,
these results demonstrate that the RPE scale offers an
acceptable method of quantifying training intensity for
collision sport athletes.
Definition of Injury
For the purpose of this study, an injury was defined as any
noncontact, soft-tissue injury sustained by a player during
a training session or match that
prevented the player from com-
pleting the entire training ses-
sion or match (17). To allow
comparisons among seasons,
injuries were also classified
according to matches missed
as a result of the injury (22). All
contact injuries were excluded
from the analysis.
Statistical Analyses
Differences in training loads
among the preseason, early-com-
petition, and late-competition
training phases and between the
2006–2007 and 2008–2009 sea-
sons were analyzed using a 2-way
(season 3training phase) analy-
sis of variance with repeated
measures. Injury exposure was
calculated by multiplying the
number of players by
the session duration. Injury in-
cidence was calculated by di-
viding the total number of injuries
by the overall injury exposure.
Influence of Training Load on
Likelihood of Injury. The training
load and injury data collected in
the first 2 seasons were mod-
eled to provide the likelihood of
soft-tissue injury with a given
training load. Individual train-
ing load and injury data were
modeled using a logistic re-
gression model with a binomial
distribution (injury vs. no in-
jury) and logit link function.
Data were analyzed in SAS
using the PROC GENMOD
procedure. This procedure was used based on its ability to
handle logistic regression. PROC GENMOD also has the
ability to handle unbalanced models (where different
numbers of repeated results were available for each player).
The summary statistic used for assessing the adequacy of the
fitted logistic model (goodness of fit) was the scaled deviance
of the model and significance of independent variables. A
scaled deviance of 1 refers to a perfect fit. Based on the
distribution of the training load data and the scaled deviance,
it was also determined that the model was best fitted with the
log of the training load per week. This model was also chosen
as an identical model had been validated and used previously
to investigate the training–injury relationship in subelite
collision sport athletes (21).
Figure 1. Weekly planned and actual training loads for the preseason, early-competition, and late-competition
training phases in elite collision sport athletes.
Figure 2. Preseason, early-competition, and late-competition training loads for the 2006–2007 and 2008–2009
seasons in elite collision sport athletes.
2596
Journal of Strength and Conditioning Research
the
TM
Injury Prediction in Collision Sport Athletes
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
Injury Prediction Model. Based on the results of the logistic
regression model, an injury prediction model encompassing
planned and actual training loads was developed and
implemented to determine if noncontact, soft-tissue injuries
could be predicted and therefore prevented. Differences in
planned and actual training loads throughout the season were
analyzed using Cohen’s effect size (ES) statistic. Effect sizes of
0.2, 0.5, and 0.8 were considered small, moderate, and large,
respectively (7). A moderate difference (i.e., ES of 0.5)
between the planned and actual training load was required
for the ÔthresholdÕtraining load to be exceeded. Effect sizes
were calculated by multiplying the between-subject SD for
training load by 0.5 (7). Given
that a relationship was identi-
fied between training load and
injury (from the first 2 seasons),
it was possible to plan training
loads to minimize the effect of
injuries. When actual training
loads for individual players
were higher (based on a mod-
erate ES) than planned training
loads, then judgments were
made on the individual man-
agement of the player (e.g.,
players continued to train with-
out modifications, or training
loads for those individual play-
ers were modified, or reduced
to minimize the risk of injury).
Injury prevalence was calcu-
lated as the proportion of play-
ers injured when actual training
loads exceeded planned train-
ing loads.
Sensitivity and Specificity of the
Injury Prediction Model. Data
were crossvalidated to determine
TABLE 2. Site and type of injuries resulting in missed matches for the 2006–2007 and 2008–2009 seasons.*†
2006–2007 Season 2008–2009 Season
Number Injury rate % Number Injury rate %
Site
Thorax/abdomen 6 0.5 (0.1–0.8) 20.0 2 0.2 (0.0–0.4) 7.7
Anterior thigh 6 0.5 (0.1–0.8) 20.0
Posterior thigh 12 0.9 (0.4–1.4) 40.0 10 0.8 (0.3–1.2) 38.5
Groin 2 0.2 (0.0–0.4) 7.7
Knee 6 0.5 (0.1–0.8) 20.0 9 0.7 (0.2–1.1) 34.6
Calf 3 0.2 (0.0–0.5) 11.5
Type
Muscular strains 24 1.8 (1.1–2.6) 80.0 17 1.3 (0.7–1.9) 65.4
Overuse 4 0.3 (0.0–0.6) 13.3 6 0.5 (0.1–0.8) 23.1
Joint sprains 2 0.2 (0.0–0.4) 6.7 3 0.2 (0.0–0.5) 11.5
Total 30 2.3 (1.5–3.1) 100.0 26 2.0 (1.2–2.8) 100.0
*The number and incidence of injuries represents any noncontact, soft-tissue injury that resulted in a missed match.
†Injury rates are expressed per 1,000 exposure hours (and 95% confidence intervals).
Figure 3. Incidence of new and recurrent injuries for the 2006–2007 and 2008–2009 seasons in elite collision
sport athletes. The incidence of injuries represents any noncontact, soft-tissue injury that resulted in a missed
match.
VOLUME 24 | NUMBER 10 | OCTOBER 2010 | 2597
Journal of Strength and Conditioning Research
the
TM
|
www.nsca-jscr.org
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
the accuracy of the injury pre-
diction model. The proportion of
true positive (i.e., predicted injury
and player sustained injury) and
negative (i.e., no injury predicted
and the player did not sustain an
injury) results, and false positive
(i.e., predicted injury but the
player did not sustain any injury)
and negative (i.e., no injury pre-
dicted but the player sustained an
injury) results were also calcu-
lated to describe errors made in
the statistical decision process
and to allow the calculation of
sensitivity (i.e., the proportion of
injured players who were pre-
dicted to be injured) and speci-
ficity (i.e., the proportion of
uninjured players who were pre-
dicted to remain injury-free) like-
lihood ratios (1). Sensitivity and
specificity values were calculated
using the following equations:
Sensitivity ¼True Positive=ðTrue Positive
þFalse NegativeÞ;
Specificity ¼True Negative=ðFalse Positive
þTrue NegativeÞ:ð1Þ
Positive and negative likelihood ratios were calculated
using the following equations:
Likelihood ratio positive ¼Sensitivity=ð1SpecificityÞ;
Likelihood ratio negative ¼ð1
SensitivityÞ=Specificity:ð1Þ
.
Although there were 91 players in the sample, injury
predictions based on the training loads performed by
individual players were made on a weekly basis, so that
within the total cohort, there was a total number of true
positive and negative predictions, and a total number of false
positive and negative predictions. Data are expressed as
means and 95% confidence intervals (CIs), and the level of
significance was set at p#0.05.
RESULTS
Training Loads
The planned and actual training loads for the preseason, early-
competition, and late-competition training phases are shown
in Figure 1. The preseason training loads were greater than
the early-competition and late-competition training loads.
Comparison of Training Loads between Seasons
No significant differences (p.0.05, ES = 0.03–0.30) were
detected in training loads for the 2006–2007 and 2008–2009
preseason, early-competition, or late-competition training
phases (Figure 2).
Figure 4. Relationships between training load, training phase, and likelihood of injury in elite collision sport athletes.
TABLE 3. Actual training loads of injured players, and the corresponding prevalence of noncontact, soft-tissue injury.*†
Preseason Early competition Late competition
Training load (units) 4,341 (4,082–4,600) 3,395 (3,297–3,493) 2,945 (2,797–3,094)
Injury prevalence (%) 72 (63–81)% 57 (47–67)% 75 (66–84)%
*Injury prevalence = proportion of players injured when actual training loads exceeded planned training loads. †Data are expressed as
means (and 95% confidence intervals).
2598
Journal of Strength and Conditioning Research
the
TM
Injury Prediction in Collision Sport Athletes
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
Incidence of Missed Match Injuries over the 4 Seasons
Site and Type of Missed Match Injuries. Players participated in
13,103 exposure hours over both the 2006–2007 and
2008–2009 seasons (for a total of 26,206 exposure hours).
Of the 80 contracted players in the 2008–2009 seasons, 49
(61.3%) sustained at least 1 noncontact, soft-tissue injury.
Of the players who sustained an injury, 35 (71.4%) sustained 2
or more injuries. The majority of missed match injuries
sustained over the 4 seasons were to the posterior thigh (0.8
[95% CI, 0.5–1.2] per 1,000 hours, 39.3%). Muscular strains
were the most common type of injury (1.6 [95% CI, 1.1–2.0]
per 1,000 hours, 73.2%) (Table 2).
Comparison of Missed Match Injuries between Seasons
The total number of missed match injuries in the 2006–2007
and 2008–2009 seasons was 30 and 25, respectively. The
incidence of soft-tissue injury resulting in a missed match was
2.3 (95% CI, 1.5–3.1) per 1,000 hours and 1.9 (95% CI, 1.2–2.7)
per 1,000 hours for the 2006–2007 and 2008–2009 seasons,
respectively. Although not significant (p.0.05), the incidence
of new injuries was greater in the 2006–2007 seasons (2.0 [95%
CI, 1.2–2.8] per 1,000 hours vs. 1.1 [95% CI, 0.6–1.7] per 1,000
hours), whereas the incidence of recurrent injuries was greater
in the 2008–2009 seasons (0.3 [95% CI, 0.0–0.6] per 1,000
hours vs. 0.8 [95% CI, 0.3–1.2] per 1,000 hours) (Figure 3).
TABLE 4. Site and type of predicted injuries during the 2008–2009 season.*†
Site Number Injury rate % Type Number Injury rate %
Thorax 2 0.2 (0.0–0.4) 1.2 Muscular strains 110 8.4 (6.8–10.0) 68.3
Lumbar 8 0.6 (0.2–1.0) 5.0 Overuse 48 3.7 (2.6–4.7) 29.8
Anterior thigh 13 1.0 (0.5–1.5) 8.1 Joint sprains 1 0.1 (0.0–0.2) 0.6
Posterior thigh 57 4.4 (3.2–5.5) 35.4
Groin 22 1.7 (1.0–2.4) 13.7
Knee 25 1.9 (1.2–2.7) 15.5
Calf 21 1.6 (0.9–2.3) 13.0
Foot 10 0.8 (0.3–1.2) 6.2
Other 1 0.1 (0.0–0.2) 0.6
Total 159 12.1 (10.2–14.0) 100.0 Total 159 12.1 (10.2–14.0) 100.0
*The number and incidence of injuries represent any noncontact, soft-tissue injury that prevented the player completing the entire
training session or match.
†Injury rates are expressed per 1,000 exposure hours (and 95% confidence intervals).
TABLE 5. Accuracy of model for predicting noncontact, soft-tissue injuries.*†
Actual status
Injured Not injured
Predicted status Predicted injury True positive False positive
N= 121 N=20
Predicted no injury False negative True negative
N=18 N= 1589
Sensitivity Specificity
87.1 (80.5–91.7)% 98.8 (98.1–99.2)%
Likelihood ratio positive 70.0 (45.1–108.8)
Likelihood ratio negative 0.1 (0.1–0.2)
*True positive = predicted injury and player sustained injury; false positive = predicted injury but player did not sustain injury; false
negative = no injury predicted but player sustained injury; true negative = no injury predicted and player did not sustain injury; sensitivity =
proportion of injured players who were predicted to be injured; specificity = proportion of uninjured players who were predicted to remain
injury-free; Likelihood ratio positive = sensitivity/(12specificity); likelihood ratio negative = (12sensitivity)/specificity.
†Although there were 91 players in the sample, injury predictions based on the training loads performed by individual players were
made on a weekly basis, so that within the total cohort, there was a total number of true positive and negative predictions, and a total
number of false positive and negative predictions. Sensitivity and specificity data and positive and negative likelihood ratios are
expressed as rates (and 95% confidence intervals).
VOLUME 24 | NUMBER 10 | OCTOBER 2010 | 2599
Journal of Strength and Conditioning Research
the
TM
|
www.nsca-jscr.org
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
Influence of Training Load on Likelihood of Injury
Based on the results of the logistic regression model, players
were 50–80% likely to sustain a preseason injury within the
training load range of 3,000–5,000 units. These training load
ÔthresholdsÕwere considerably reduced (1,700–3,000 units) in
the late-competition phase of the season (Figure 4). The
actual preseason, early-competition, and late-competition
training loads performed by injured players were 4,341 (95%
CI, 4,082–4,600), 3,395 (95% CI, 3,297–3,493), and 2,945
(95% CI, 2,797–3,094) units, respectively. When actual
training loads for individual players were higher than
planned training loads, noncontact, soft-tissue injury oc-
curred in 72 (95% CI, 63–81), 57 (95% CI, 47–67), and 75
(95% CI, 66–84)% of cases, for the preseason, early-
competition, and late-competition training phases, respec-
tively (Table 3).
Injury Prediction Model
Site and Type of Predicted Injuries. A total of 159 noncontact,
soft-tissue injuries were sustained over the latter 2 seasons
(2008–2009). The majority of injuries predicted in the
2008–2009 seasons were to the posterior thigh (4.4 [95%
CI, 3.2–5.5] per 1,000 hours, 35.4%). Muscular strains were the
most common type of injury (8.4 [95% CI, 6.8–10.0] per 1,000
hours, 68.3%) (Table 4).
Sensitivity and Specificity of the Injury Prediction Model
The percentage of true positive predictions was 62.3% (n=
121), whereas the total number of false positive and false
negative predictions was 20 and 18, respectively. The sensi-
tivity and specificity of the injury prediction model were 87.1
(95% CI, 80.5–91.7)% and 98.8 (95% CI, 98.1–99.2)%,
respectively (Table 5). The calculated positive and negative
likelihood ratios were 70.0 (95% CI, 45.1–108.8) and 0.1 (0.1–0.2),
respectively.
DISCUSSION
The present study is the first to describe an injury prediction
model for noncontact, soft-tissue injuries in elite collision
sport athletes. The use of the injury prediction model allowed
medical staff to make informed decisions on training
management on a scientific basis, whereas modeling of
training load and injury data identified the training load
ranges that resulted in an ‘‘acceptable’’ injury risk and also the
training load ranges that resulted in an ‘‘unacceptable’’ injury
risk. These findings provide a scientific method of monitoring
and regulating training load that removes the ÔguessworkÕ
from training.
Previous studies of collision sport athletes have reported
a significant relationship (r= 0.86) between training loads
and training-injury rates (17), suggesting that the harder these
athletes train, the more injuries they will sustain. Further-
more, reductions in training loads have been shown to reduce
training-injury rates and result in greater improvements in
_
VO
2
max (18). However, it has also been shown that collision
sport athletes that perform ,18 weeks of preseason training
before sustaining an initial injury are at increased risk of
sustaining a subsequent injury, whereas players with a low
off-season
_
VO
2
max are at increased risk of sustaining a contact
injury (20). Clearly, training for collision sports reflects
a balance between the minimum training load required to
elicit an improvement in fitness and the maximum training
load tolerable before sustaining marked increases in injury
rates. In the elite team sport environment, it is critical to have
the maximum number of players free from injury and
available for selection in as many games as possible
throughout the season (34,36). The developed injury
prediction model provides a practical framework to monitor
training loads and prevent noncontact, soft-tissue injuries in
elite collision sport athletes.
Modeling of training load and injury data of players
identified safe and unsafe training loads for different phases of
the season. In the preseason training period, players were
50–80% likely to sustain an injury within the training load
range of 3,000–5,000 units. However during the late-
competition phase, significantly less training could be
tolerated before increasing the likelihood of injury; players
were 50–80% likely to sustain an injury within the training
range of 1,700–3,000 units. In all 3 training phases (preseason,
early-competition, and late-competition), small increases in
training load resulted in large increases in injury likelihood.
Indeed, when actual training loads for individual players were
higher than planned training loads, noncontact, soft-tissue
injury occurred in 72, 57, and 75% of cases, for the preseason,
early-competition, and late-competition training phases,
respectively. Collectively, these results give practical in-
formation on the training–injury dose–response relationship
in elite collision sport athletes.
Players that exceeded the training load threshold were
70 times more likely to test positive for noncontact, soft-tissue
injury, whereas players that did not exceed the training load
threshold were injured 1/10 as often. These results demon-
strate that the developed model was adequately sensitive to
detect both players who were susceptible to noncontact, soft-
tissue injuries and players who were not susceptible to injury.
It should be emphasized that in 62.3% of cases, a player was
highlighted as having the potential for injury and no
intervention was undertaken (and the player was sub-
sequently injured). These results compare favorably with
the proportion of players (11.3%) who sustained a soft-tissue
injury despite not being predicted as being at risk for soft-
tissue injury. Although there are numerous examples of highly
skilled individuals using ÔintuitionÕto identify potential
problems within their area of expertise (25), the present
findings suggests that the information generated from
training monitoring could have been used more efficiently.
Despite the additional scientific information generated by the
injury prediction model, strength and conditioning coaches
were more comfortable using their intuitive ÔexpertiseÕto
manage the training loads of players. Unfortunately, the
model had far greater accuracy predicting injuries (62.3%)
2600
Journal of Strength and Conditioning Research
the
TM
Injury Prediction in Collision Sport Athletes
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
than it did influencing the decision making of strength and
conditioning staff, with ,14% of players provided alternative
training or additional recovery when an injury was predicted.
It is likely that a combination of intuitive ÔexpertiseÕand the
injury prediction model may have provided greater accuracy
in predicting training-related, soft-tissue injuries (38,40).
Given the success in predicting injuries and that the total
proportion of incorrect predictions was small (23.9%), these
results suggest that the injury prediction model provides
greater sensitivity than the sole judgment of strength and
conditioning staff.
No significant differences existed between the initial 2
seasons (where no injury prediction model was used) and
latter 2 seasons (where the injury prediction model was
developed and implemented) for the incidence of missed
match injuries. These findings may be attributed to the similar
training loads employed in the initial and latter seasons.
However, it is also possible that the failure to reduce
noncontact, soft-tissue injuries with the introduction of the
injury prediction model was because of the low overall
incidence of missed match soft-tissue injuries over this period.
Indeed, the soft-tissue injury rates reported in this study are
considerably lower than other running-based team sports
(e.g., Australian football) (35). It is therefore likely that
a training monitoring model designed to capture training
load-related soft-tissue injuries may have better application to
team sports with a greater emphasis on running, and a lower
emphasis on physical collisions. Indeed, team sports such as
Australian football, basketball, and soccer, which typically
employ high running loads to condition players, would likely
benefit more from the developed injury prediction model
than collision sports that have a greater emphasis on
conditioning through physical contact (e.g., American
football). Nonetheless, the present findings demonstrate
the applicability of an injury prediction model to detect
noncontact, soft-tissue injuries in elite collision sports that
incorporate high volumes and intensities of both physical
contact and running into their conditioning programs.
It should be noted that the injury prediction model has
a number of potential limitations. Firstly, the developed
model is unable to predict collision injuries. However,
although collision injuries are multifactorial, it is likely that
players with better developed physical qualities may be less
susceptible to collision injuries, whereas overtrained players
may be at a greater risk of injury. Indeed, several rugby league
studies have reported a significant relationship (r= 0.86)
between training loads and training-injury rates (17),
suggesting that the harder these athletes train, the more
injuries they will sustain. Furthermore, although limited
evidence exists in elite collision sport athletes, previous
studies on subelite collision sport athletes have shown that
performing ,18 weeks of preseason training before sustain-
ing an initial injury increases the risk of sustaining a sub-
sequent injury, whereas players with a low off-season
_
VO
2
max are at an increased risk of sustaining a contact
injury (20). Secondly, the accuracy of the model for
predicting training load-related soft-tissue injuries is de-
pendent on the quality of data entered. In the present study,
training loads were estimated from the session-RPE, a sub-
jective measurement of training load. Future injury prediction
models could include global positioning system data, data on
the number and intensity of collisions, markers of fatigue
(e.g., self-reports, neuromuscular and hormonal data), and
match work rates to further refine the present training
monitoring system. Finally, although the model was success-
ful in predicting injuries over 2 competitive seasons, its
continued success is dependent on knowledge of planned
and actual training loads. Planned training loads for
conditioning, strength, and skill sessions are required for
the injury prediction model to operate effectively.
Although the injury prediction model used in this study had
sufficient predictive accuracy to warrant systematic use in an
elite team sport program, a fine balance exists between
training, detraining, and overtraining. Furthermore, although
a relationship was observed between training load and
likelihood of injury, training programs must also be
physiologically efficient and psychologically appropriate to
allow players to cope with the demands of competition (6).
Indeed, exposing the brain to hard physical work and fatigue
on a regular basis appears to improve the body’s ability to
cope with fatigue; physically intense training not only
improves physical fitness but equally importantly also
increases the mental durability of players (33). Physically
(and mentally) unfit players are more likely to pace
themselves as a self-preservation and protection strategy
(33). If players have not been exposed to hard physical work
on a regular basis, the brain instructs the body to stop
exercise earlier to prevent exhaustion (33). With this in mind,
it may be argued that it is worthwhile prescribing high
training loads (note, not excessive) to players to determine
which players are most susceptible to injury under physically
stressful situations (these players most likely will not tolerate
the intensity and fatigue of competition), and which players
are not susceptible to injury under physically stressful
situations (these players are more likely to tolerate the
intensity and fatigue of competition).
In conclusion, this study developed an injury prediction
model for noncontact, soft-tissue injuries in elite collision
sport athletes. Players that exceeded the training load
threshold were 70 times more likely to test positive for
noncontact, soft-tissue injury, whereas players that did not
exceed the training load threshold were injured 1/10 as often.
Modeling of training load and injury data identified the
training load ranges that resulted in an ‘‘acceptable’’ injury
risk and also the training load ranges that resulted in an
‘‘unacceptable’’ injury risk. From a practical perspective,
these findings provide information on the training dose–
response relationship, and a scientific method of moni-
toring and regulating training load in elite collision sport
athletes.
VOLUME 24 | NUMBER 10 | OCTOBER 2010 | 2601
Journal of Strength and Conditioning Research
the
TM
|
www.nsca-jscr.org
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
PRACTICAL APPLICATIONS
There are several practical applications of this study that are
relevant to the strength and conditioning coach. Monitoring
and regulating training loads is critical to ensure that players
receive a progressively overloaded periodized training pro-
gram and are given adequate recovery between high-volume
and high-intensity sessions. It is important for strength and
conditioning coaches to work closely with sport scientists to
determine the appropriate training and recovery periods to
maximize improvements in performance without unduly
increasing injury incidence.
Players that exceeded the training load threshold were
70 times more likely to test positive for noncontact, soft-tissue
injury, whereas players that did not exceed the training load
threshold were injured 1/10 as often. In 62.3% of cases,
a player was highlighted as having the potential for injury, and
no intervention was undertaken (and the player was
subsequently injured). Given the success in predicting injuries,
and that only a small proportion (23.9%) of incorrect
predictions were made, these results suggest that the injury
prediction model provides greater sensitivity than the sole
judgment of strength and conditioning staff. Indeed, a large
proportion of injuries may have been prevented if strength
and conditioning staff heeded the warnings provided through
the scientific analysis of the training loads.
Finally, although the injury prediction model provided
a successful framework to manage noncontact, soft-tissue
injuries it should be recognized that the model was based on
planned and actual training loads. An assumption with this
model is that players possess adequate physical qualities to
perform rigorous physical training and that the planned
training loads were adequate to develop and maintain
physical fitness. To some extent, this framework constrains
the amount of physical adaptation permitted through
training, by limiting the amount of physical work that can
be performed. Allowing players to exert themselves above
and beyond the planned training loads may result in soft-
tissue injury but could also produce greater physical
adaptations and mental durability in players.
REFERENCES
1. Altman, DG and Bland, JM. Diagnostic tests. Sensitivity and
specificity. BMJ 308: 1552, 1994.
2. Avalos, M, Hellard, P, and Chatard, JC. Modeling the training-
performance relationship using a mixed model in elite swimmers.
Med Sci Sports Exerc 35: 838–846, 2003.
3. Bannister, EW and Calvert, TW. Planning for future performance:
Implications for long term training. Can J Appl Sports Sci 5: 170–176,
1980.
4. Bannister, EW, Calvert, TW, Savage, MV, and Bach, T. A systems
model of training for athletic performance. Aust J Sports Med 7:
57–61, 1975.
5. Bannister, EW, Good, P, Holman, G, and Hamilton, CL. Modeling
the training response in athletes. In: The 1984 Olympic Scientific
Congress Proceedings. Sport and Elite Performers. (Vol. 3), Landers, DM,
ed. Champaign, IL: Human Kinetics, 1986. pp. 7–23.
6. Baron, B, Moullan, F, Deruelle, F, and Noakes, TD. The role of
emotions on pacing strategies and performance in middle and long
duration sport events. Br J Sports Med, in press.
7. Cohen, J. Statistical Power Analysis for the Behavioral Sciences (2nd ed.).
Hillsdale, NJ: Lawrence Erlbaum, 1988.
8. Dunbar, CC, Robertson, RJ, Baun, R,Blandin, MF, Metz, K,Burdett, R,
and Goss, FL. The validity of regulating exerciseintensity by ratings of
perceived exertion. Med Sci Sports Exerc 24: 94–99, 1992.
9. Filaire, E, Lac, G, and Pequignot, JM. Biological, hormonal, and
psychological parameters in professional soccer players throughout
a competitive season. Percept Mot Skills 97: 1061–1072, 2003.
10. Foster, C. Monitoring training in athletes with reference to
overtraining syndrome. Med Sci Sports Exerc 30: 1164–1168, 1998.
11. Foster, C, Daines, E, Hector, L, Snyder, AC, and Welsh, R. Athletic
performance in relation to training load. WMJ 95: 370–374, 1996.
12. Foster, C, Daniels JT, and Yarbrough, RA. Physiological and training
correlates of marathon running performance. Aust J Sports Med 9:
58–61, 1977.
13. Foster, C, Florhaug, JA, Franklin, J, Gottschall, L, Hrovatin, LA,
Parker, S, Doleshall, P, and Dodge, C. A new approach to
monitoring exercise training. J Strength Cond Res 15: 109–115, 2001.
14. Gabbett, TJ. Severity and cost of injuries in amateur rugby league: A
case study. J Sports Sci 19: 341–347, 2001.
15. Gabbett, TJ. Incidence of injury in semi-professional rugby league
players. Br J Sports Med 37: 36–43, 2003.
16. Gabbett, TJ. Influence of injuries on team playing performance in
rugby league. J Sci Med Sport 7: 340–346, 2004.
17. Gabbett, TJ. Influence of training and match intensity on injuries in
rugby league. J Sports Sci 22: 409–417, 2004.
18. Gabbett, TJ. Reductions in pre-season training loads reduce training
injury rates in rugby league players. Br J Sports Med 38: 743–749, 2004.
19. Gabbett, TJ. Changes in physiological and anthropometric charac-
teristics of rugby league players during a competitive season.
J Strength Cond Res 19: 400–408, 2005.
20. Gabbett, TJ and Domrow, N. Risk factors for injury in sub-elite
rugby league players. Am J Sports Med 33: 428–434, 2005.
21. Gabbett, TJ and Domrow, N. Relationships between training load,
injury, and fitness in sub-elite collision sport athletes. J Sports Sci 25:
1507–1519, 2007.
22. Gibbs, N. Injuries in professional rugby league. A three-year
prospective study of the South Sydney professional rugby league
football club. Am J Sports Med 21: 696–700, 1993.
23. Gissane, C, Jennings, D, Kerr, K, and White, J. Injury rates in rugby
league football: Impact of change in playing season. Am J Sports Med
31: 954–958, 2003.
24. Gissane, C, Jennings, DC, Cumine, AJ, Stephenson, SE, and White, JA.
Differences in the incidence of injury between rugby league forwards
and backs. Aust J Sci Med Sport 29: 91–94, 1997.
25. Gladwell, M. Blink. The Power of Thinking without Thinking.
New York, NY: Little, Brown and Company, 2005.
26. Hoffman, JR, Kang, J, Ratamess, NA, and Faigenbaum, AD.
Biochemical and hormonal responses during an intercollegiate
football season. Med Sci Sports Exerc 37: 1237–1241, 2005.
27. Impellizzeri, FM, Rampinini, E, Coutts, AJ, Sassi, A, and Marcora, M.
Use of RPE-based training load in soccer. Med Sci Sports Exerc 36:
1042–1047, 2004.
28. Krebs, PS, Zinkgraf, S, and Virgilio, SJ. Predicting competitive
bicycling performance with training and physiological variables.
J Sports Med Phys Fitness 26: 323–330, 1986.
29. Morton, RH. Modelling training and overtraining. J Sports Sci 15:
335–340, 1997.
30. Morton, RH, Fitz-Clarke, JR, and Bannister, EW. Modeling human
performance in running. J Appl Physiol 69: 1171–1177, 1990.
2602
Journal of Strength and Conditioning Research
the
TM
Injury Prediction in Collision Sport Athletes
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
31. Mujika,I,Busso,T,LaCoste,L,Barale,F,Geyssant,A,andChatard,JC.
Modeled responses to training and taper in competitive swimmers. Med
Sci Sports Exerc 28: 251–258, 1996.
32. Mujika,I,Chatard,JC,Busso,T,Geyssant,A,Barale,F,andLaCoste,L.
Effects of training on performance in competitive swimming. Can J Appl
Physiol 20: 395–406, 1995.
33. Noakes, TD, St. Clair Gibson, A, and Lambert, EV. From
catastrophe to complexity: A novel model of integrative central
neural regulation of effort and fatigue during exercise in humans:
summary and conclusions. Br J Sports Med 39: 120–124, 2005.
34. Orchard, J. Occasional piece: On the value of team medical staff–can
the ÔMoneyballÕapproach be applied to injuries in professional
football? Br J Sports Med 43: 963–965, 2009.
35. Orchard, J and Seward, H. Epidemiology of injuries in the Australian
football league, seasons 1997–2000. Br J Sports Med 36: 39–44, 2002.
36. Piggott, B, Newton, MJ, and McGuigan, MR. The relationship
between training load and incidence of injury and illness over a pre-
season at an Australian football club. J Aust Strength Cond 17: 4–17,
2009.
37. Scrimgeour, AG, Noakes, TD, Adams, B, and Myburgh, K. The
influence of weekly training distance on fractional utilization of
maximum aerobic capacity in marathon and ultramarathon runners.
Eur J Appl Physiol 55: 202–209, 1986.
38. Shanks, B. Scout’s honor. The bravest way to build a winning team.
New York, NY: Sterling and Ross, 1999.
39. Stewart, AM and Hopkins, WG. Seasonal training and performance
of competitive swimmers. J Sports Sci 18: 873–884, 2000.
40. Surowiecki, J. The wisdom of crowds: Why the many are smarter
than the few and how collective wisdom shapes business, economics,
societies and nations. New York, NY: Anchor Books, 2004.
VOLUME 24 | NUMBER 10 | OCTOBER 2010 | 2603
Journal of Strength and Conditioning Research
the
TM
|
www.nsca-jscr.org
Copyright © National Strength and Conditioning Association Unauthorized reproduction of this article is prohibited.
