Introduction
The impact of injuries on athletes varies from minor pain with no detrimental effect
on their ability to train and compete to short-term, long-term, career-ending and
catastrophic injuries [1] and long-term health
conditions such as osteoarthritis [2]. Injuries in
sport can be managed effectively using a risk management approach [3]
[4] with the key stages
of this approach being risk identification, risk assessment, risk evaluation and
risk mitigation [3]
[5].
Injury surveillance studies underpin the assessment stage of the risk management
process and provide the evidence-base on which risk evaluations are undertaken and
cost-effective injury prevention strategies are developed and justified [6]. Injury burden, which is a composite measure of
injury incidence and mean injury severity [7], is an
important parameter in several areas of sport, as the data are relevant to
athletes’ short and long-term health [8] and
their ability to train and compete [9]
[10]. Consensus statements for sports injury
surveillance studies are well established and the recommended procedures and output
measures enable injury burden values to be calculated in both athlete welfare [11]
[12]
[13] and athlete performance [9]
[10] contexts. Although injury burden has
been recognised and reported as an output measure from injury surveillance studies
in rugby for over 20 years [14]
[15]
[16], the importance
of injury burden information has only more recently been recognised and recommended
for other sports [17].
The principles, practices and problems associated with recording and using injury
burden values have been discussed previously [7]
[18]. The increasing importance given to reporting
injury burden values in injury surveillance studies has, however, been accompanied
by errors and misunderstandings about how to calculate, present and interpret the
data. For injury burden measurements to deliver their full range of benefits, it is
essential that valid results are reported in order to avoid reaching incorrect
conclusions and making invalid recommendations.
The objective of this review is to explain why median severity and ordinal severity
scales should not be used to calculate and report injury burden results in injury
surveillance studies.
Materials and Methods
A range of parameters is used to characterise and compare injuries sustained in
sport. Typically, these parameters include injury location, type, nature
(e. g. acute, gradual onset) and cause (e. g. contact, non-contact,
competition, training) [11]
[12]
[13]. The primary output measures
reported in injury surveillance studies are (i) the frequency with which injuries
are sustained (incidence) and (ii) a central-tendency value for the severity of
injuries sustained. Injury severity in a sample population may be reported as the
mode, median or mean value, all expressed as the number of days-absence resulting
from injury. If the severity data follow a normal distribution the three
central-tendency measures return the same value [19].
Sports injury severity values, however, rarely follow a normal distribution and,
while it is theoretically possible for injury severity data to be left-skewed, they
most often follow a right-skewed distribution [18]
[20]
[21]
[22]. This means that the three
central-tendency values do not have the same value and their values will usually
follow the sequence of mean>median>mode. Severity values can also be
recorded using ordinal scales, such as those recommended in some sports injury
surveillance consensus statements [12]
[13].
Injuries that result in athletes being unable to train or compete are referred to
as
‘time-loss injuries’; those that do not prevent an athlete from
training and competing are referred to as ‘non-time-loss injuries’
[12]
[13]. Injury
burden, which describes the total number of days that an athlete population is
unable to train or compete, can be expressed as the total days-absence resulting
from injuries sustained in a specified setting over a specified period of time. For
non-time-loss injuries that might impact on an athlete’s ability to train
and/or compete at their normal performance level, days-absence can be
replaced by the number of full-time-equivalent (FTE) days-absence based on the
magnitude of the effect the injuries have on the athletes’ activities [9]
[10]. For example, an
athlete only able to train at 80% of their normal performance level for 10
days would be classified as having sustained {10 x (100–80)}/100 FTE
days-absence, equal to 2 FTE days. Because the total number of days-absence recorded
in a study is dependent on the sample size and the period of exposure, injury burden
values are usually normalised and reported as the number of
days-absence/1000 athlete-hours [7]
[18]. Normalised injury burden values can be calculated
in two ways: (i) as [total days-absence x 1000]/[total athlete-hours of
exposure] or (ii) as [injury incidence (expressed as the number of
injuries/1000 athlete-hours) x mean severity of injury (expressed as
days-absence)]. Injury burden reported as days-absence/1000 athlete-hours of
exposure is, therefore, directly related to the total number of days-absence, with
the slope of the relationship equal to [1000/total athlete-hours of
exposure]. Plotting mean injury severity values against injury incidence values
within a risk matrix containing iso-risk contours provides a simple, visual method
to present injury burden results and to highlight the highest and lowest-risk
injuries in a particular setting [7]
[18].
Results
[Table 1] presents a lower limb injury data set for 25
(hypothetical) lower limb injuries sustained by one rugby team playing 25 games (15
players/game; 80 minutes/game; total exposure: 500
player-hours): these data equate to an overall incidence of lower limb injuries in
the sample population of 50.0 injuries/1000 player-hours
(25×1000/500). [Table 1] shows the
number of player-days-absence associated with each of the 25 injuries together with
the incidences and the mean and median severities of the injuries sustained at each
injury location. In addition, injury burden values are presented for each injury
location based on the incidence values and both the mean and median injury severity
values for the locations. [Fig. 1] displays these two
sets of injury burden values plotted against the corresponding total days-absence,
for the five body locations. [Fig. 2] presents a risk
matrix with the mean and median severities of injury plotted against the
corresponding incidence values for the 5 injury locations (see [Table 1] for the data used).
Fig. 1 Relationships between injury burdens, for 5 injury locations,
calculated using mean and median severity values and the total days-absence
shown for the injury locations in [Table
1].
Fig. 2 Risk matrix showing mean and median injury severity values
plotted against corresponding incidence values for the 5 lower limb injury
locations (based on data shown in [Table 1]).
The risk contours represent injury burden values of 25, 100 and 250
days-absence/1000 player-hours; (● mean injury
severity values; ▲: median injury severity values).
Table 1 Hypothetical lower limb injury data set for a rugby
team playing 25 games (15 players/game) of 80 minutes
duration (total match exposure: 500 player-match-hours).
|
Injury location (number of injuries); days-absence
|
|
Thigh (7)
|
Ankle (6)
|
Foot (5)
|
Groin (4)
|
Knee (3)
|
|
1
|
2
|
10
|
2
|
4
|
3
|
|
2
|
3
|
12
|
4
|
7
|
7
|
|
3
|
6
|
16
|
7
|
10
|
70
|
|
4
|
12
|
24
|
40
|
35
|
|
|
5
|
20
|
43
|
75
|
|
|
|
6
|
25
|
55
|
|
|
|
|
7
|
35
|
|
|
|
|
|
Match exposure (player-hours)
|
500
|
500
|
500
|
500
|
500
|
|
Total days-absence (days)
|
103
|
160
|
128
|
56
|
80
|
|
Incidence(i)
|
14.0
|
12.0
|
10.0
|
8.0
|
6.0
|
|
Mean severity, days-absence
|
14.7
|
26.7
|
25.6
|
14.0
|
26.7
|
|
Median severity, days-absence
|
12.0
|
20.0
|
7.0
|
8.5
|
7.0
|
|
Burden (mean severity) (ii)
|
206
|
320
|
256
|
112
|
160
|
|
Burden (median severity) (ii)
|
168
|
240
|
70
|
68
|
42
|
(i): Incidence reported as injuries/1000 player-hours; (ii): Burden
reported as days-absence/1000 player-hours.
Incidence, mean and median severities and injury burden values based on previously
published rugby data [17], are presented for eight
body locations (lumbar spine, hip/groin, wrist/hand, chest, lower
leg, foot, ankle, thigh) in [Table 2].
Table 2 Incidence, median severity, mean severity and injury
burden based on median and mean severity of injuries sustained by rugby
players at eight body locations. Data derived from Table 6 in [17].
|
Injury location
|
Incidence;
|
Injury severity, days
|
Injury burden; days-absence/1000 player-hours
|
|
injuries/1000 player-hours
|
Median
|
Mean
|
Based on median severity
|
Based on mean severity
|
|
Lumbar spine
|
1.5
|
10
|
44.0
|
15
|
66
|
|
Hip/groin
|
1.9
|
9
|
43.2
|
17
|
82
|
|
Wrist/hand
|
4.5
|
10
|
43.1
|
45
|
194
|
|
Chest
|
3.8
|
13
|
19.7
|
49
|
75
|
|
Lower leg
|
4.0
|
17
|
47.5
|
68
|
190
|
|
Foot
|
1.9
|
37
|
44.2
|
70
|
84
|
|
Ankle
|
6.9
|
15
|
46.4
|
104
|
320
|
|
Thigh
|
6.4
|
14
|
26.7
|
90
|
171
|
Four examples of ordinal scales used to record the consequences of injuries in
clinical and performance-related contexts are shown in [Table 3]. While [Fig. 3] illustrates the
consistency between these ordinal scales in terms of their rank order, it highlights
the non-linearity of the scales in terms of days-absence. [Table 4] shows the injury severity data presented in [Table 1] after the days-absence values have been
transformed into data based on the ordinal scales shown in [Fig. 3] and columns 1 and 2 of [Table 3].
[Table 4] also includes the corresponding mean
and median severity values derived from these ordinal scale values and the
corresponding injury burden values based on the mean and median ordinal severity
scale values. [Fig. 4] plots the two sets of injury
burden data shown in [Table 4] against the
corresponding total days-absence values for the five body locations.
Fig. 3 Relationships between ordinal consequence scales (see [Table 3]) and days-absence from injury.
Fig. 4 Relationships between total days-absence and injury burden
values calculated using mean and median ordinal scale severity values (see
[Table 4]).
Table 3 Examples of ordinal scales used for recording injury
consequences in clinical and performance-related contexts.
|
Arbitrary scale values
|
Clinical context
|
Performance context
|
|
Grouped days-absence
|
Treatment/rehabilitation
|
|
0
|
No days-absence
|
No treatment or rehabilitation required
|
No reduction in athletic performance
|
|
3
|
1 to 7 days-absence
|
Minor treatment and/or rehabilitation required
|
Minor reduction in athletic performance
|
|
6
|
8 to 28 days-absence
|
Moderate treatment and/or rehabilitation required
|
Moderate reduction in athletic performance
|
|
9
|
>28 days-absence
|
Major treatment and/or rehabilitation required
|
Major reduction in athletic performance
|
Table 4 Injury data presented in [Table 1] after severity values have been transformed into
ordinal scale values (based on the scales shown in [Table 3], columns 1 and 2).
|
Injury location (number of injuries); ordinal scale severity
values (i)
|
|
Thigh (7)
|
Ankle (6)
|
Foot (5)
|
Groin (4)
|
Knee (3)
|
|
1
|
3
|
6
|
3
|
3
|
3
|
|
2
|
3
|
6
|
3
|
3
|
3
|
|
3
|
3
|
6
|
3
|
6
|
9
|
|
4
|
6
|
6
|
9
|
9
|
|
|
5
|
6
|
9
|
9
|
|
|
|
6
|
6
|
9
|
|
|
|
|
7
|
9
|
|
|
|
|
|
Match exposure (player-hours) (ii)
|
500
|
500
|
500
|
500
|
500
|
|
Total days-absence (days) (ii)
|
103
|
160
|
128
|
56
|
80
|
|
Incidence (iii)
|
14.0
|
12.0
|
10.0
|
8.0
|
6.0
|
|
Mean severity (ordinal scale) (iv)
|
5.1
|
7.0
|
5.4
|
5.3
|
5.0
|
|
Median severity (ordinal scale) (iv)
|
6.0
|
6.0
|
3.0
|
4.5
|
3.0
|
|
Burden (mean severity) (v)
|
71
|
84
|
54
|
42
|
30
|
|
Burden (median severity) (v)
|
84
|
72
|
30
|
36
|
18
|
(i): See [Table 2], columns 1 and 2; (ii): See
[Table 1] for exposure value and total
days-absence; (iii): Incidence reported as injuries/1000
player-hours; (iv): Severity derived from the ordinal scale values ([Table 2]); (v): Burden based on the incidence
of injury and the mean and median severity ordinal scale values.
Discussion
One-off injury burden measurements are normally used to report the level of injury
risk experienced by a defined athlete population during specific competitions,
events or over defined periods of time. Injury burden values can also be ranked as
a
function of, for example, injury location, type or causation to create injury risk
spectra in order to identify priorities for injury prevention. Repeated,
longitudinal measurements of injury burden are used to monitor long-term trends in
injury risk or to measure injury risks pre- and post-interventions such as injury
prevention strategies. Injury burden values can be displayed visually in risk
matrices by plotting mean severity values against the corresponding injury incidence
values [7]
[18].
Whichever reporting method is adopted, however, it is essential that the data are
calculated and presented correctly.
Because the three central-tendency values of severity are rarely the same in injury
surveillance studies, it is essential that the correct severity value (i. e.
the mean value) is used to calculate injury burden. Instead of using mean severity
of injury to calculate and present injury burden results, some publications have
used the median severity of injury [20]
[21]
[22]. The first
section of this review, therefore, explains why median severity values should not
be
used to calculate or present injury burden results. The results presented in [Table 1] and [Fig. 1]
show that injury burden results based on the median severity of injury
under-estimate the true injury burden. [Fig. 1] also
demonstrates that, while a linear relationship exists between total days-absence and
injury burden values based on the mean severity of injury, no such linear
relationship exists when using the median severity of injury. Hence, using median
severity values to calculate and present injury burden results can be seen to have
no mathematical validity will inevitably give rise to misleading conclusions and
recommendations.
If median injury severity values are used, instead of mean severity values, to
present injury burden data in a risk matrix, the resultant graphs generate incorrect
risk profiles, which will again lead to misleading conclusions and recommendations;
see [Table 1] and [Fig.
2]. In this case, the injury burden results derived from the mean severity
values place two body locations (ankle, foot) in the ‘ high-risk
region’ of the risk matrix, three body locations (thigh, knee, groin) in the
‘medium-risk region’ and no body locations in the ‘low-risk
region’. If median injury severity values are used, no body locations are
placed in the ‘high-risk region’ of the matrix, two injuries (ankle,
thigh) in the ‘medium-risk region’ and three injuries (foot, groin,
knee) in the ‘low-risk region’. Of the five body locations, only the
results for the thigh appear in the same injury risk region (medium-risk) when using
both the mean and median severity-based injury burden results.
The results presented in [Table 2] demonstrate that
the conclusions presented above are not merely an inherent factor of the
hypothetical data set used in this review. Reviewing the injury burden results based
on median severity ([Table 2], column 5) would lead
to the following incorrect conclusions:
-
Injury burden values for lumbar spine (15) and hip/groin (17)
injuries are similar;
-
Injury burden values for wrist/hand (45) and chest (49) injuries are
similar;
-
Injury burden values for lower leg (68) and foot (70) injuries are similar;
and
-
Injury burden values for ankle (104) and thigh (90) injuries are similar.
If these injuries are compared correctly using injury burden values based on
mean severity ([Table 2], column 6), it can
be seen that the correct conclusions are:
-
Injury burden for hip/groin injuries (82) is ~25%
higher than that for lumbar spine injuries (66);
-
Injury burden for wrist/hand injuries (194) is almost three times
higher than that for chest injuries (75);
-
Injury burden for lower leg injuries (190) is more than twice that for foot
injuries (84);
-
Injury burden for ankle injuries (320) is almost twice that for thigh
injuries (171).
A further source of error arises when injury burden values based on median severity
values are presented and assessed in risk matrices containing iso-risk contours
based on mean injury severity values. Comparing the injury burden results shown in
[Table 1] and [Fig.
2] based on mean severity leads to the following conclusion regarding the
rank order for the risks of injury:
Ankle>Foot>Thigh>Knee>Groin. Using injury burden
values based on the median severity of injuries, however, leads to a different
conclusion regarding the rank order of injury risks:
Ankle>Thigh>Foot>Groin>Knee.
The second section of this review explains why ordinal scales of injury severity
should not be used to calculate or present injury burden results. While [Fig. 1] confirmed the expected linear relationship
between total days-absence and injury burden values based on mean severity using
days-absence, [Fig. 4] shows that linear
relationships do not exist between total days-absence and injury burden values when
both mean and median severity values derived from ordinal severity scales are used.
As injury burden provides a ratio scale of injury risk, the injury incidence and
injury severity values used to calculate injury burden must both be based on ratio
scales. The requirements for ratio scales are that they include an absolute zero and
equal interval scale values: these criteria enable ratio scale values to be added,
subtracted, multiplied and divided [19]. Injury
severity values based on days-absence meet the criteria for a ratio scale and these
values can, therefore, be averaged to provide mean injury severity values. Injury
severity values derived from ordinal scales, which do not have equal scale values
(see [Table 3] and [Fig.
3]), can be used for ranking purposes but the values cannot be added,
subtracted, multiplied or divided [19]. Ordinal scale
severity values cannot, therefore, be used to calculate meaningful injury burden
values. Although applying numerical scale values, such as 0, 3, 6, 9, to ordinal
scales, (see [Table 3], column 1) may (i) give an
illusion that the intervals are equal and (ii) create the impression that
mathematical calculations can be carried out using the values, there is no
mathematical justification for this.
[Fig. 3] highlights the non-linearity of ordinal
scales and the disproportionality of the scale intervals in terms of days-absence.
Ordinal scale severity values should, therefore, not be used in injury surveillance
studies for calculating injury burden results, as they introduce the potential for
errors in the conclusions and recommendations presented. For example, from the
ordinal scales presented in [Table 3] and [Fig. 3], it can be seen that if one athlete in an
injury surveillance study sustained a minor injury requiring 2 days-absence (ordinal
severity scale value: 3) and a second athlete in the study sustained a moderate
injury requiring 8 days-absence (ordinal severity scale value: 6), the total
time-loss from athletic activity would be 10 days (mean severity: 5 days-absence).
These two injuries, however, would lead to a mean severity score in the study of 4.5
if based on the arbitrary ordinal scale. If in a second study, one athlete sustained
an injury that did not result in time-loss (ordinal severity scale value: 0) and a
second athlete in the study sustained an injury that resulted in 100 days-absence
(ordinal severity scale value: 9), the total time loss from athletic activity would
be 100 days (mean severity: 50 days-absence). These two injuries would, however,
also lead to a mean severity score of 4.5 when based on the ordinal severity scale.
Therefore, presenting the injury burden results from these studies based on ordinal
scale severity values implies that the injury burden outcomes from the two studies
were identical, even though the actual injury burden in the second study was
ten-times higher than the injury burden in the first study.
Using ordinal scale mean and median severity values both lead to the incorrect
ranking of injury risks. For example, the results presented in [Table 4] show that injury burden derived from mean
values of the ordinal scales rank the risks as:
Ankle>Thigh>Foot>Groin>Knee, while injury burden
values derived from median values of the ordinal scale rank the risks as:
Thigh>Ankle>Groin>Foot>Knee. Neither of these rank
orders of injury burden agrees with the correct evaluation based on mean
days-absence, which is shown above.
When injury consequences are assessed in a performance-related context, injury
consequences are often recorded on a daily or weekly basis as a
‘daily/weekly severity score’. These individual
daily/weekly ordinal scale values are summed to produce a ‘total
severity score’ for the duration of the athlete’s adverse health
condition. The ‘total severity score’ values recorded for all cases
of the adverse health condition are then averaged to produce a ‘mean
severity score’ for the adverse health condition. These ‘mean
severity scores’ are then often plotted against the incidences of the
adverse health conditions within a risk matrix. The same arguments regarding the
non-validity of using ordinal scale values that were discussed above also apply in
these cases. Furthermore, this approach is compounded further if several ordinal
scale values are used to provide a number of injury severity outcome measures, which
are then summed to derive a ‘daily/weekly severity score’
for an adverse health condition.
The discussion and examples presented above illustrate why ordinal scale severity
values produce incorrect injury burden values and why these will in turn also lead
to incorrect and inconsistent conclusions and recommendations from injury
surveillance studies.
Injury burden data derived from injury surveillance studies provide important
information about athletes’ risks of injury; this information enables injury
risks to be quantified and evaluated and injury prevention priorities to be
identified [7]
[18].
Unfortunately, in many of these studies, the principles associated with collecting,
calculating, reporting and presenting injury burden data have not been fully
understood. As a consequence, median severity values and/or ordinal scales
of severity have been incorrectly employed to calculate and present injury burden
values in a range of sports, including: football [20]
[21]
[22],
ice hockey [23]
[24]
[25], athletics [26]
[27], gymnastics [28] and
Paralympics [29]
[30].
The discussion and examples presented here have demonstrated the nature of the errors
associated with using median severity and ordinal severity scales to calculate and
present injury burden results. It is recommended that researchers intending to
publish injury burden results are familiar with the correct procedures for
calculating, reporting and presenting injury burden results in order to avoid
presenting incorrect injury burden results, conclusions and recommendations.
