ScienceDirect - Physical Therapy in Sport : Motor control and strength as predictors of hamstring injury in elite players of Australian football

Matt Cameron ^,^,^a, Roger Adams^b and Christopher Maher^b

^a Physiotherapist, Sydney Swans Football Club, P.O. Box 173, Paddington, NSW 2021, Australia
^b School of Physiotherapy, University of Sydney, Sydney, Australia

Available online 11 July 2003.

1. Introduction

Hamstring strain injuries are common in all football codes and sports involving sprinting, and are the most frequently occurring and recurring of all injuries in Australian football. At the elite level, hamstring injuries occur at a rate of 6.2 injuries per club per season, and result in 21.2 missed player games per club per season ([Orchard and Seward, 2002]). These rates are the highest of all the elite level football codes in Australia ( [Seward et al., 1993]). Hamstring injuries have the highest recurrence rate of all football injuries and notwithstanding the best rehabilitation attempts, more than one in three (34%) injuries recur within the same season ( [Orchard and Seward, 2002]). Risk has been shown to be increased following previous hamstring injury ( [Bennell et al., 1998, Garrett, 1996 and Orchard, 2001]), calf strain injury ( [Orchard, 2001]), and serious knee and groin injuries ( [Verrall et al., 2001]). Risk has also been shown to be increased in players of Australian football older than 23 years of age ( [Orchard, 2001]).

Despite the magnitude of this problem in Australian football, the aetiology of such injuries is unclear. Factors that have been suggested to predispose an athlete to hamstring muscle strain injury include muscle weakness, muscle imbalance, poor flexibility, fatigue, inadequate warm-up, poor neuromuscular control and poor running technique ([Agre, 1985]). There is little evidence for poor flexibility as a hamstring injury predictor ( [Bennell et al., 1999]), and apart from muscle weakness there is no empirical support for any of the other suggested factors ( [Orchard, 2001]).

The hamstring muscle group reaches peak elongation and acts eccentrically at the hip and knee during the late swing phase of the running cycle ([Frigo et al., 1979 and Simonsen et al., 1985]). Kinetic and EMG studies reveal that the hamstrings are most active and develop the greatest torques at the hip and knee during late swing through to the mid-stance phase of running ( [Mann and Sprague, 1980 and Montgomery et al., 1994]). It is during these parts of the running cycle that the hamstrings are under the greatest demands and injury most likely ( [Mann and Sprague, 1980 and Stanton and Purdam, 1989]). Given the high forces involved, it would seem that hamstring weakness might predispose an athlete to injury, however, to date, there have not been adequate findings to support either hamstring muscle weakness or hamstring-quadriceps strength imbalance as risk factors ( [Orchard, 2001]).

Muscle weakness has been the most extensively investigated of all proposed predisposing factors for hamstring injury. Retrospective studies have suggested a relationship between muscle weakness and hamstring injury ([Crosier and Crielaard, 2000, Crosier et al., 2002 and Heiser et al., 1984]), however, retrospectivity can confound the causes and effects of injury. The literature also contains a small number of prospective studies, which allow firmer conclusions to be drawn regarding the factors involved in the prediction of injury, but the results to date are inconsistent. Two studies using isometric hamstring strength assessment demonstrated an association between injury and a side-to-side deficit of 10% ( [Burkett, 1970]) and a lower hamstring-to-quadriceps ratio ( [Yamamoto, 1993]). The latter finding was supported by a study using isokinetic dynamometry at 60 °/s, which linked injury to a hamstring-to-quadriceps ratio below 0.60 ([Orchard et al., 1997]). In contrast, [Liemohn, 1978] did not find any relationship between injury and isometric hamstring-to-quadriceps strength ratio, nor did a larger study examining concentric and eccentric isokinetic strength reveal any association between injury and similar strength ratios ( [Bennell et al., 1998]).

One possible causative factor of hamstring injury, that has not been previously examined, is the accuracy of neuromuscular control of the leg during running. Throughout the running cycle there are many challenging neuromuscular events in which the hamstring acts, for example, to control hip and knee motion in late swing and to provide hip extensor torque in early stance. During sprinting, these events occur over a very short period of time, and if the control and coordination are inadequate, then muscle strain injury may result ([Agre, 1985 and Bennell et al., 1999]). Control of a limb requires that information be obtained and integrated from proprioceptors of the entire limb, and that control will be influenced in part by activity in the opposite limb. Joint or single segment proprioception has been assessed by various techniques that employ joint position testing, kinesthaesia testing or sense of effort testing, often in non-weight-bearing postures ( [Lephart and Fu, 2000]). [Waddington and Adams, 1999] assessed movement discrimination (MD) at the ankle based on functional movement principles, whereby subjects performed ankle inversion movements in standing and made judgments regarding the extent of these movements without visual input about the movement. This requires the processing of both afferent and efferent information about the lower limb being tested, and performance reflects a subject's ability to do this accurately. An association was demonstrated between poor discrimination ability and previous ankle sprain injury. This task was later extended to enable assessment of MD of the knee during weight-bearing flexion ( [Waddington et al., 2000]).

Utilising the same functional movement principles for the current study on hamstring function, a similar apparatus may be employed to assess the discrimination of movement extent using the backward swinging leg, whilst weight-bearing on the other side, in order to create a functional movement as close as possible to the movement at injury. The position of the player while being tested is selected to recreate the proposed movement region in which hamstring injury occurs: between the late swing to mid-stance phases of the running cycle (Fig. 1). Ability to accurately discriminate swinging leg movements can be seen as reflecting use of proprioception from the lower limbs, and is integral to the motor control of the leg in this region of the running cycle where hamstring injury is likely to occur.

(7K)
Fig. 1. Left leg late swing phase of the running cycle.

The aim of this study was to assess swinging leg MD, isokinetic hamstring and quadriceps strength, and history of previous hamstring injury, in order to determine any association these factors have with respect to subsequent hamstring injuries in a group of elite Australian football players.

2. Method

2.1. Participants

Twenty players of Australian football were recruited for this study from the training squad of one professional Australian Football League (AFL) team. All subjects were male and the mean (SD) age was 23.6 years (3.2), height 185.5 cms (8.5), weight 87.8 kg (9.1), and the mean number of AFL training years was 4.7 (3.34). Subjects were excluded if any significant lower limb injury was sustained in the twelve weeks immediately prior to assessment. Approval for the study was obtained from the University of Sydney Human Ethics Committee, and all subjects signed an informed consent.

2.2. Tests

2.2.1. Movement discrimination testing protocol

Lower limb motor control was assessed with a purpose-built active movement extent discrimination apparatus, the AMEDA ([Waddington and Adams, 1999]). This apparatus consists of a vertical contact plate attached to the end of the motor-driven shaft, which sets the stop for ending leg swinging movements at different positions. A laptop computer controlled the Programmable Stepper Motor ¹ driving the shaft, with end-point re-positioning accuracy manufacturer-specified at 0.01 mm. Five stop settings were used to generate the five leg swing movement extents, and these were each 5.6 mm apart. The stop positions were located from 20–43 mm behind the posterior margin of the heels in standing, which corresponded to an angle of the lower limb to vertical of approximately 2 ° and represented the position of the leg at early to mid-stance phase of running (see Fig. 2).

(6K)
Fig. 2. A subject standing on the active movement discrimination apparatus and swinging the right leg to contact the start bar (A) and then the stop plate (B).

Testing of leg swing movement commenced with subjects in normal weight-bearing stance astride the apparatus, heads up and eyes focused on a point on the wall opposite, so that they had no direct vision of their foot or the contact plate. The contact plate was set to one of the five positions and on command, subjects transferred weight to one leg, actively lifted the tested leg to touch the start bar, before swinging the limb backward toward the plate until contact was made, and then returning to standing (Fig. 2). After this movement the subject identified which one of the five stop settings (i.e. movement extents) they had just experienced.

Each subject was given a series of trial leg swing movements on the AMEDA in order to familiarize them with the feel of each stop setting prior to data collection. Following this, all of the five stop positions were presented eight times in random order. Subjects were allowed only one leg swing movement for each trial. After each movement the subject was asked which of the five stop numbers they felt corresponded to the movement extent performed. The reported stop position was recorded for the forty trials, and no feedback as to reporting accuracy was provided. Both legs of each subject were tested and the side first tested was randomly determined. From this testing, a measure of a subject's ability to differentiate between different movement extents was obtained.

2.2.2. Muscle strength testing protocol

Strength measurements were conducted on all players in the pre-season of 2000 at the Sport Science Department of the New South Wales Academy of Sport, University of Sydney, Australia. Knee flexion and extension concentric strength were assessed with a Cybex II Isokinetic Dynamometer^,² using a protocol which most players had experienced at previous screening sessions. Warm-up prior to data collection consisted of five minutes cycling on a stationary bike and stretching of the lower limb musculature. For isokinetic testing, subjects were positioned in sitting with straps around the thigh, waist and chest, with arms folded across the chest. The axis of rotation of the dynamometer was aligned with the centre of the lateral femoral condyle as outlined by [Perrin, 1993]. The shin pad of the dynamometer shaft was placed 2 cm proximal to the lateral malleolus. After set-up, subjects completed several submaximal knee flexion/extension trials. Torque and displacement were determined at an angular velocity of 60 °/s, on both legs, and for both knee flexion and extension. At each direction, subjects were required to exert maximum effort for three repetitions, and the mean of the two highest torque values was recorded. A minimum two-minute recovery period was imposed between repeated trials. The order of leg testing was randomised, however, the quadriceps was tested prior to the hamstrings.

The Cybex II dynamometer was computer-interfaced and torque and angular displacement data were collected for 5 s at a sampling rate of 1000 Hz. This information was stored to disk for later processing, which included the determination of peak torque that in all instances occurred after the impact spike ([Kannus, 1994]). Given this protocol, ramping and damping procedures were considered unnecessary. Prior to, and after testing, the torque output of the Cybex was calibrated by measuring a range of known torques according to the manufacturer's specification. Peak torque values were normalised for body weight for each subject to allow comparison of knee flexion and extension concentric strength between subjects.

2.2.3. Procedure

Previous hamstring injury occurrence in the two years prior to the study was determined by questionnaire and review of club injury records, and an injury was deemed to be significant if any missed matches resulted. All subjects were tested prior to the 2000 football season, and monitored throughout the 2000 and 2001 seasons for significant hamstring injury. The club medical officer made a diagnosis of significant hamstring injury upon satisfying all of the following criteria

(a)acute onset of pain in the posterior thigh during training or a match
(b)posterior thigh pain reproduced with stretch, contraction and palpation of the hamstring muscle
(c)hamstring muscle injury demonstrated by magnetic resonance imaging (MRI)
(d)at least one match missed due to hamstring injury

The medical officer was blind to the results of MD and isokinetic testing.

2.2.4. Statistical analysis

The reported stop positions obtained from discrimination testing were cast into stimulus-response matrices, and non-parametric signal detection analysis was used to obtain MD scores, defined as the area under the Receiving Operating Characteristic (ROC) curves ([McNicol, 1972]). ROC curves were used to plot the probability of correctly identifying the extreme stop positions against the probability of incorrectly labeling them as other stop positions. The area under a ROC curve provides an overall measure of stop position discrimination ability in this task, and was calculated with the ROC subroutine in SPSS 10.0 for Windows. An area of 0.5 corresponds to chance whereas 1.0 corresponds to perfect discriminability ( [Maher and Adams, 1995]). From the strength data, peak concentric torque of the hamstring and quadriceps muscle groups were used in data analysis, and a hamstring-to-quadriceps strength ratio (H/Q) was calculated for each leg of each subject. These data were analysed with SPSS 10.0 for Windows software. Prospective and retrospective injured/uninjured grouping structures were created by the occurrence of hamstring injuries in the two-year period following measurement, and by the past history of hamstring injuries in the two years preceding measurement. Thus determined, separate ANOVAs were conducted on the muscle strength and MD variables, with factors Injury Status (injured/uninjured) and Side (dominant/non-dominant).

Previous injury, MD, and hamstrings and quadriceps muscle strength variables were evaluated by the use of ROC curves as to their performance of predicting prospective injury status grouping. Use of ROC curves, originally developed to measure the ability of a subject to differentiate between different stimuli ([Green and Swets, 1966]), has been extended to medical research, where the curves have been used to examine a diagnostic test's discriminative capability for determining presence of disease or injury ( [Swets et al., 2000]). Applied in this way, curves were generated by plotting the true-positive rate (sensitivity) and false-positive rate (1-specificity) along the vertical and horizontal axes for each of the predictor variables. The area under the ROC curve is the most useful comparative measure of the performance of screening tests ( [Hanley and McNeil, 1982 and Park et al., 2002]), and a value over 0.8 can be interpreted as a test with good predictive power ( [Meijer et al., 2002]). Areas under the ROC curve were calculated using SPSS 10.0 for Windows, and comparisons between variables and calculations of cut-off points were performed with Prediction program version 3.0.

3. Results

In the two seasons following testing, six subjects sustained one or more significant hamstring muscle strains. In the two seasons prior to testing, seven subjects had experienced hamstring muscle strains, and two of these subjects were in the group of six injured in the subsequent period. There was no past history of hamstring injury in 4 of the 6 subsequently injured subjects.

Mean MD scores and thigh concentric strength variables for the groups are given in Table 1. With respect to the prospective analyses, MD scores were significantly worse in those subsequently injured compared to the subsequently uninjured group (F(1,18)=9.44,p=0.007). No player with a MD score above the group mean subsequently injured a hamstring, but six of the ten players with scores below the mean did incur injury. Subsequently injured subjects also had significantly lower hamstrings-to-quads (H/Q) strength ratios (F(1,18)=8.75,p=0.008) and significantly greater quadriceps strength adjusted for their body weight (F(1,18)=6.13,p=0.02) than uninjured subjects. Hamstring muscle strength values, however, showed no significant differences between groups (F(1,18)=0.83,p=0.37).

Table 1. Mean (SD) movement discrimination score (MD), relative peak torques (N m/kg) and hamstring-to-quadriceps (H/Q) ratio for previously and subsequently injured and uninjured players

When the group was separated into previously injured and uninjured subjects, there were no significant differences on any of the variables. No differences were found between dominant and non-dominant sides in either the prospective or retrospective analyses, and there were no interactions between side and injury status. Because there are strong temporal and spatial relationships between the lower limbs during locomotion ([Shapiro et al., 1981]), indicating a high level of inter-limb coordination, the two legs during running have been considered a single coordinative structure ( [Sherwood, 1989]). Accordingly, the two limbs for each subject were averaged to give a single score for MD and for each strength assessment. These form the data presented in Table 1. To examine the injury-predictive capability of the tested variables, mean MD score, hamstring and quadriceps strength, and H/Q ratio across limbs were calculated for each player. The ability of each variable to predict hamstring injury was evaluated by calculating the area under the relevant ROC curve. These values, level of significance and confidence intervals are presented in Table 2.

Table 2. Area under the ROC curve (Area), standard error (SE), asymptotic significance (p) and 95% confidence interval (CI) for hamstring injury predictor variables: movement discrimination score (MD), relative peak hamstring (H) and quadriceps (Q) torque, hamstring-to-quadriceps ratio (H/Q), and previous hamstring injury (Prev Inj)

Movement discrimination score, quadriceps strength and H/Q ratio were significant predictors of hamstring injury based on the area under the ROC curve. Pair-wise comparisons revealed no significant difference between MD score and quadriceps strength in terms of their value as predictors (z=0.3278,p=0.3715) or H/Q ratio (z=0.1038,p=0.4587), or between quadriceps strength and H/Q ratio (z=0.4107,p=0.3406). There was no significance difference between hamstring and quadriceps strength (z=1.1230,p=0.1307) as injury predictors, however, there was a significant difference in hamstring injury predictive value between hamstring strength and H/Q ratio (z=2.8558,p=0.0021) and between hamstring strength and MD score (z=1.9295,p=0.0268). Hamstring strength and previous hamstring injury were individually not significant predictors of subsequent hamstring injury, and there were no significant differences between them. The sample size was too small to permit the development of an equation combining the predictor variables.

To develop cut-points or decision thresholds ([Swets et al., 2000]) for the significant hamstring injury predictor variables, sensitivity and specificity values and a summary measure, Youden's index, were calculated for each variable, and are presented in Table 3. Youden's index is the best summary measure of a diagnostic test's ability ([Biggerstaff, 2000]), and is calculated for a given cut-point by adding the sensitivity and specificity values and subtracting one.

Table 3. Sensitivity (SENS), specificity (SPEC), and Youden's index (YI) for movement discrimination (MD), hamstring-to-quadriceps strength ratio (H/Q) and relative quadriceps (Q) peak torque (Nm/kg) cut-point values

4. Discussion

Three measures—backward leg swing MD, hamstring strength relative to quadriceps strength and quadriceps strength relative to body weight—were found to predict hamstring injury in a subsequent two-season period. The number of players sustaining a hamstring injury in this study was equivalent to a seasonal rate of 15% and is similar to the rate of 14% described by a previous study of injuries in Australian football at this level ([Seward et al., 1993]).

First, on a test utilizing a movement similar to the action at the most likely time of injury, subsequently hamstring-injured players had below average MD ability, and this would suggest that it is players with poor lower limb proprioception and motor control who are at risk of hamstring injury. If an error is made in the control of the swinging lower leg at a time in the running cycle when high hamstring tissue forces exist, then a strain injury is possible.

The performance of swinging leg MD testing as a screening tool for hamstring injury was assessed using ROC curve analysis. Movement discrimination testing is a good predictor of subsequent hamstring injury, as indicated by obtaining an area under the ROC curve of 0.869. A range of MD scores were assessed as the cut-point or decision threshold for this measure, whereby players scoring below this point are deemed to have a positive test and predicted to sustain a hamstring injury, and scores above this point are deemed negative and not predicted to injure. Consequently, sensitivity and specificity values were calculated for each decision threshold and are listed in Table 3. The Youden's index for each MD score acting as the decision threshold is also indicated in Table 3. A MD score of 0.78 corresponds to the decision threshold with the highest Youden's index of 0.71, having 100% sensitivity and 71.4% specificity. These values suggest that the MD test is better at identifying those players unlikely to injure, and that some players testing positive are able to avoid injury for other reasons. As with any predictive test, the cut-point can be varied according to the requirements of the clinician. In professional football where hamstring injuries are one of the most frequently occurring and recurring injuries, and players do have the opportunity to engage in injury prevention programs, the identification of any player at risk is worthwhile. In this situation, the cost of false positives is less than the cost of false negatives, (i.e. not identifying players who subsequently sustain a hamstring injury). However, leg swing MD testing lacks clinical usefulness without an intervention program to rectify any deficit, and this is a direction for future research.

In relation to the second significant injury predictor, over the two-year study period, those players who sustained a hamstring injury also had a lower H/Q strength ratio when measured using isokinetic dynamometry at 60 °/s. The magnitude of the difference in H/Q ratio between injured and uninjured groups is similar to a previous study of elite footballers ([Orchard et al., 1997]). However, from the third significant injury predictor identified, it would seem that it is the increased quadriceps strength rather than a decrease in hamstring strength that is responsible for the reduced H/Q ratio in the injured group of this study. This is a finding that has not been reported previously. The increased quadriceps strength of injured subjects suggests that footballers who develop excessive quadriceps strength increase the risk of hamstring injury, despite otherwise adequate hamstring strength. The hamstring and quadriceps muscles co-contract during the early stance phase of running as the knee flexes after ground contact then extends. If quadriceps force development during this co-contraction is in excess of the hamstring muscle group's capacity, then hamstring injury may be possible. There is further co-contraction later in stance and a shift from hip extension and knee flexion torques, to hip flexion and knee extension dominance in mid-stance ( [Mann and Sprague, 1980]). It is at this point that, [Yamamoto, 1993] suggests an imbalance between the hamstrings and the hip flexor rectus femoris may result in injury. The development of additional quadriceps strength may result in an athlete running at a speed or with a technique that predisposes the insufficient hamstring muscle to injury ( [Muckle, 1982]).

Players of Australian football would be aware of the common and recurring nature of hamstring injuries, and may be purposefully increasing their hamstring strength training. This may involve increasing the number of lower limb strength exercises generally, and as a consequence, the quadriceps muscles may be strengthened to a point that places the hamstrings at further injury risk, despite any previous strength gains.

From the data here, low H/Q ratio and excessive quadriceps strength are two strength factors that are associated with an increased hamstring injury risk in Australian football. Both factors can be considered to be good predictors of hamstring injury with areas under the ROC curve of 0.881 and 0.827, respectively. Table 3 contains the sensitivity and specificity values, and Youden's index for the corresponding H/Q ratios found in this study. The traditional H/Q ratio of 0.60, when used as a cut-point for identifying at-risk players, results in this test variable as having 50% sensitivity, 85.7% specificity and Youden's index of 0.36. If 0.60 is used, half of the hamstring-injured players would be missed, however, if the cut-point is raised to 0.66 then sensitivity increases to 100%, specificity falls to 71.4%, and Youden's index nearly doubles to 0.71. None of the players in this study with a H/Q ratio above 0.66 sustained a hamstring injury in the two years subsequent to testing. These findings suggest that a higher H/Q ratio of 0.66 should be used as the threshold for detecting elite players of Australian football at-risk of hamstring injury.

The hamstring injury risk associated with high quadriceps strength relative to body weight poses an interesting clinical dilemma in professional football. Intervention to reduce quadriceps strength may involve reducing leg weight training, however, this may have performance repercussions and be undesirable in a large proportion of players. A high quadriceps strength cut-off will decrease the number of players involved in any intervention program that may harm performance, however, it will increase the number of injuries missed i.e. false negatives. Considering the consequences of injury and intervention, the relationship between excessive quadriceps strength and hamstring injury deserves to be further investigated.

A ratio of eccentric hamstring and concentric isokinetic strength has been suggested as useful in assessment of hamstring-injured subjects ([Aagaard et al., 1995 and Crosier and Crielaard, 2000]). A significant difference in such a ratio in previously hamstring-injured athletes was demonstrated in a retrospective investigation ( [Crosier et al., 2002]), but not in a prospective study ( [Bennell et al., 1998]). There are no prospectively designed studies supporting an eccentric/concentric ratio identifying hamstring injury risk athletes, however, there is with a concentric only ratio ( [Orchard et al., 1997]). The current study was also limited to only concentric strength and ratios, however, these findings suggest that the early stance phase of running involving concentric hamstring activity, rather than the eccentric late swing phase, may be a more likely region of hamstring injury. The role of eccentric strength and the region of the running cycle of hamstring injury deserve further investigation.

Eccentric strength assessment involves some injury risk, particularly in hamstring muscles ([Orchard et al., 2001]), therefore a concentric thigh strength assessment is recommended for the pre-season screening of hamstring injury risk in participants of Australian football. The motor control test did not place any stress on capacity to generate force, on range, or on speed, but it did challenge the capacity of the players to make fine discriminations between different extents of backward leg swinging movements. The motor control measure does not produce muscle pain or fatigue, is less disruptive to training, and is a useful screening tool for hamstring injury risk at any time of the season.

The results of this study support an association between hamstring injury and both a low MD score and a high degree of hamstring and quadriceps strength imbalance. [Crosier et al., 2002] argue that the relationship between muscle imbalance and injury has always been a logical assumption, however, we would argue that the relationship between poor accuracy in relevant movement control and hamstring injury is equally logical. One possible explanation for the pattern of predictors observed here is that weight training to develop leg strength can improve quadriceps strength relatively more easily than hamstring strength, yet without the level of motor control needed for injury-free performance with a stronger system. One implication is that there may exist an interaction between thigh muscle strength and leg swing MD such that a deficit in one variable can be compensated by a higher ability in the other. The decision threshold for each variable and any possible interaction need to be validated in a subsequent group of footballers. Future research should also be directed at training methods that can reliably improve MD in backward leg swing, in order to reduce the risk of hamstring injury.

5. Conclusion

The findings of this study suggest that poor leg neuromuscular control may be a significant contributor to hamstring injury. Data in this study has also supported the injury risk of a low hamstring-to-quadriceps ratio. Investigations into only a single contributing factor are likely to lack agreement due to the multi-factorial nature of hamstring strain injury. This study has linked two variables—poor leg MD and thigh muscle strength imbalance—with an increased risk of hamstring injury. At this point, it is recommended that players be screened for hamstring injury risk with leg motor control testing and thigh muscle strength measurements, so that changes obtained in these values could be used as a basis for designing effective injury prevention programs.

Motor control and strength as predictors of hamstring injury in elite players of Australian football

Abstract

Article Outline