- Email: [email protected]
Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index
G Model
GAIPOS-4384; No. of Pages 6 Gait & Posture xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Gait & Posture journal homepage: www.elsevier.com/locate/gaitpost
Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index Morgan Sangeux a,b,c,*, Jill Rodda a,b,c, H. Kerr Graham a,b,c a
The Royal Children’s Hospital, Australia Murdoch Childrens Research Institute, Australia c The University of Melbourne, Australia b
A R T I C L E I N F O
A B S T R A C T
Article history: Received 1 September 2014 Received in revised form 23 December 2014 Accepted 30 December 2014
The identification of gait patterns in cerebral palsy offers a common language for clinicians and contributes to management algorithms. We describe a quantitative classification of sagittal gait patterns based on the plantarflexor–knee extension couple index. This consists of a scatter plot based on ankle and knee scores, and allows objective identification of the sagittal gait pattern. Sagittal kinematic data from 200 limbs of 100 patients with bilateral spastic cerebral palsy were utilized to validate the algorithm against the assessment of a clinician with expertise in gait pattern identification. A dataset of 776 cerebral palsy patients, 1552 limbs, was used to compare the sagittal gait patterns against k-means statistical clustering. The classification was further explored with respect to the knee kinetics during the middle of stance and physical examination measurements of the gastrocnemius–soleus complex. Two supplementary materials (Appendices 2 and 3) provide in-depth discussion about statistical properties of the plantarflexor–knee extension couple index as well as its relationship with statistical clustering. The plantarflexor–knee extension index achieved 98% accuracy and may be suitable for the computational classification of large patient cohorts and multicentre studies. The sagittal gait patterns were strongly related to k-means statistical clustering and physical examination of the gastrocnemius– soleus complex. Patients in crouch gait had normal soleus and gastrocnemius lengths but spasticity in the gastrocnemius. Patients in jump gait exhibited a short gastrocnemius and soleus and gastrocnemius spasticity. Patients in true equinus presented with a moderately contracted soleus and gastrocnemius and gastrocnemius spasticity. Patients in apparent equinus did not show abnormal physical examination measurements for the gastrocnemius–soleus complex. ß 2015 Elsevier B.V. All rights reserved.
Keywords: Gait Pattern Index Physical examination Cerebral palsy
1. Introduction Davids et al. [1] described five sources of data to guide clinical decision-making for children with cerebral palsy. One of these, instrumented gait analysis, provides detailed information on the kinematics and kinetics of the joints of the lower limb. A typical instrumented gait analysis entails, for each limb, to analyze at least ten kinematics and kinetics curves. The interpretation of this data requires linking kinematic deviations with physical examination measurements, to define the gait impairments. The amount of data
* Corresponding author at: The Hugh Williamson Gait Analysis Laboratory, The Royal Children’s Hospital, 50 Flemington Road, Parkville, VIC 3052, Australia. Tel.: +61 416910735. E-mail address: [email protected] (M. Sangeux).
is invaluable to determine the appropriate treatment for a specific patient but may make it difficult to identify patterns and the construction of management algorithms. Rodda et al. [2,3], described a semi-quantitative sagittal gait pattern classification for patients with bilateral spastic cerebral palsy (BSCP). That classification combined pattern recognition with quantitative kinematic data, and was based on an extension of earlier work by Rang et al. [4], Sutherland and Davids [5] and Miller et al. [6]. The Rodda classification described five groups: crouch gait, jump gait, apparent equinus, true equinus and mild gait (within normal limits in the sagittal plane). The classification in five groups applies to the limb and the asymmetric group is introduced when the two limbs of the same patient belong to two different classifications. Based on the above sagittal gait patterns Rodda et al. derived a management algorithm which specifies the dominant muscle groups to be targeted for treatment of spasticity
http://dx.doi.org/10.1016/j.gaitpost.2014.12.019 0966-6362/ß 2015 Elsevier B.V. All rights reserved.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
2
or contracture and includes prescription of orthotics. This work has often been utilized to characterize the gait patterns of cohorts of patients (e.g. [7–11]). Although the classification was based in part on quantitative data, it is also in part qualitative and subjective. It therefore requires the involvement of a clinician who is expert in gait analysis and in clinical assessment. This requirement may restrict the use of the Rodda classification to expert clinicians and may prohibit its application in large patient cohorts. Therefore the first purpose of this study was to propose and validate an algorithm to classify Rodda’s sagittal gait patterns on a purely quantitative basis. Quantitative algorithms based on clustering of the kinematic data have been proposed previously (e.g. [12–14]). These algorithms present optimal statistical properties but may be difficult to apply clinically and they have seldom been used in management algorithms. The sagittal gait classification by Rodda was not derived from statistical clustering but from years of clinical observations and the underlying statistical properties are unknown. Therefore the second purpose of this study was to compare the Rodda sagittal gait pattern classification with statistical clustering. Sagittal plane kinematics at the ankle and knee in stance are mainly determined by the plantarflexion–knee extension couple [15]. The couple refers to the action of the gastrocnemius and soleus muscles, the ankle plantarflexors, to control both the advancement of the tibia over the foot and the knee kinetics in mid-stance. Spasticity or contracture of the gastrocnemius–soleus muscles in children with cerebral palsy is frequently present and will influence sagittal kinematics and kinetics. The third purpose of this study was to compare physical examination measurements of the plantarflexors with the sagittal gait patterns. 2. Material and methods 2.1. The plantarflexor–knee extension couple index The plantarflexor–knee extension (PFKE) index calculates the distance of the patient’s ankle and knee kinematics in mid-stance
from normative data. The period of the gait cycle used to calculate the PFKE index is set between 20 and 45% of the gait cycle (see Appendix 1, in supplementary material, for a discussion about this choice). During this period, the knee extends while the ankle dorsiflexes, the knee moment changes from an extensor moment to a flexor moment allowing the quadriceps to cease contracting and the ankle to absorb power through the eccentric contraction of the gastrocnemius–soleus complex. The PFKE index consists of two scores from the ankle and the knee. It is calculated as follow: 45 X kci mci 1 PFKEc ¼ 45 20 þ 1 i¼20 s ci where superscript c denotes the knee or ankle curve, subscript i denotes the time instant (in % of the gait cycle), ki is the value of the kinematic curve at the i% instant of the gait cycle, mi is the value of the normal kinematic curve at the i% instant and si is the value of the standard deviation from the normal kinematics curve at the i% instant. A PFKE index of (3,2) means that, on average between 20 and 45% of the gait cycle, the ankle curve is three standard deviation below normal and the knee curve is two standard deviations above normal. The normal kinematics curves originated from 35 typically developed children (17 girls, 18 boys) with an age range of 6–17 years old. Classification is based on the +1 or 1 values of the PFKE ankle and knee scores and is displayed on the PFKE scatter plot (Fig. 1 and Appendix 2 in the supplementary material). We define dPFKE as the minimum 1 D distance of one point to the other gait patterns. A large value for dPFKE means the point is located far from all the other gait classifications and therefore specific to the gait pattern it belongs to. Validation of the classification algorithm was performed by comparing the classification by the algorithm to classification by a clinical expert (J.R.) in a study of 200 lower limbs, in 100 patients with BSCP. Kinematic data from patients diagnosed with BSCP were randomly selected from our database. Accuracy of the classifier was expressed as the percentage of true classifications divided by the number of limbs assessed.
Fig. 1. The plantarflexor–knee extension scatter plot, with the ankle score in X and knee score in Y, for N = 200 limbs. The lines, corresponding to the 1 or 1 value for the ankle and knee scores, separate the sagittal gait patterns. The central section of the PFKE plot corresponds to the within normal limit (WNL) or mild sagittal gait pattern. An example for a PFKE index of (3,2) is provided. This point corresponds to the jump gait pattern and is at a minimum distance dPFKE = 1 to the nearest gait pattern (true equinus). The validation dataset is overlayed on the plot. Points were coloured in green if the automatic and expert classifications agreed and red if they disagreed. It can be noticed than only 4 points were misclassified and those were very close to the line between two classifications. This is confirmed by an average dPFKE for those 4 points of 0.2. Eight points were located in areas of the plot without a labelled gait pattern. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
2.2. Relationship with statistical classification A larger dataset was used for the remaining of this study. All patients with a diagnosis of cerebral palsy for which kinematics and kinetics gait data captured barefoot between January 2004 and December 2013 were included. Previous surgery, including plantarflexor lengthening was not an exclusion criterion for this study. Seven hundred and seventy six patients, 1552 limbs, were included. k-means clustering was chosen as the statistical classification algorithm and the methods described by Rozumalski et al. [13] were followed. Readers are referred to the original publication for specific details. Briefly, one column vector was constructed from the pelvis, hip, knee and ankle sagittal kinematics of each limb. Kinematic data were sampled for each percent of the gait cycle which lead to a vector of 4 curves 101 points = 404 points. Column vectors from the 1552 limbs were adjoined to form a 404 1552 matrix, KD. A reduced order approximation was obtained by projecting KD on the four orthogonal axes (eigenvectors or gait features [13,16]) which accounted for 96% of the total variance. k-means clustering requires predefining the number of clusters. Since the sagittal gait pattern classification has five groups, the number k of clusters was set to five. Calculations were performed in the statistical software R version 3.0.3 [17]. To compare the two classifications, a contingency table (or cross tabulation) was utilized. The contingency table C has five rows, one for each sagittal gait pattern, and five columns, one for each k-means clusters. Each element, Cij, of C corresponds to the number of limbs belonging to both the sagittal gait pattern i and the cluster j. Pearson’s x2 test, applied to C, tested for the independence between the two classifications. 2.3. Knee kinetic and physical examination of the gastrocnemius and soleus muscles The average sagittal knee moment between 20 and 45% of the gait cycle was calculated and compared between sagittal gait patterns and normative knee kinetic data. One way ANOVA and Tukey’s post-hoc group comparison was performed in Minitab (Minitab Inc., USA). Physical examination measurements relating to the lengths of the gastrocnemius and soleus muscles as well as gastrocnemius spasticity were extracted from the database. Soleus length was defined as the maximal ankle dorsiflexion angle, subject supine with the knee flexed at 908 [18]. Gastrocsoleus length was expressed as the maximal ankle dorsiflexion, subject supine with the knee extended (Angle_KneeExt [18]). Gastrocnemius spasticity was assessed using the Tardieu test, with the patient supine and the knee extended [19]. The ankle was dorsiflexed rapidly to elicit a ‘‘catch’’ which was then measured using a goniometer (Angle_Tardieu). To grade spasticity, we used the arithmetic difference Angle_KneeExt – Angle_Tardieu. An ordinal scale was fitted to the raw data with five categories: normal (0), slightly abnormal (1), abnormal (2), very abnormal (4) and extremely abnormal (5). Table 1 presents the angle ranges that define the scale. These values were derived from [20] and clinical experience. We utilized correspondence analysis, a form of correlation analysis when data are nominal, to study the association between physical examination measurements and sagittal gait patterns. Correspondence analysis allows the best possible graphical representation of contingency tables in a plane. All calculations were performed in R with the package FactoMineR [21].
3
Table 1 Relationship between the scale (ordinal nominal) and physical examination measurements angles at the ankle. The first column presents values for the soleus measurement (knee flexed at 908) and the second column values for the gastrocsoleus measurement (knee extended). The third column presents values for the difference between the gastrocsoleus measurement (static, knee extended) and the Tardieu measurement (dynamic, knee extended). Subjects are considered within normal limits (scale = 0) for: a soleus measurement greater than 258 dorsiflexion, a gastrocsoleus measurement greater than 158 dorsiflexion and a difference between the static and the dynamic measurements less than 58. Description
Normal Slightly abnormal Abnormal Very abnormal Extremely abnormal
Scale
Physical examination measurements Ankle angle (8) knee flexed at 908
Ankle angle (8) knee extended
Ankle angle difference (8) static–dynamic
0 1
25 < x 10 < x < 25
15 < x 0 < x < 15
x<5 5 < x < 10
2 3 4
0 < x < 10 10 < x < 0 Less than 10
10 < x < 0 20 < x < 10 Less than 20
10 < x < 20 20 < x < 30 30 and over
3. Results 3.1. Accuracy of the automatic classification based on the PFKE index Fig. 1 presents the four false classifications leading to 4/200 = 98% accuracy. Patients presented a continuum of deviation from normal at the knee and ankle rather than well delineated groups in separated portions of the graph. The average dPFKE for misclassified patients was 0.2. The PFKE index for eight limbs were located in sections of the plot that do not correspond to any previous sagittal gait pattern classification. These correspond to the ankle within normal limits and the knee in recurvatum or the knee within normal limits and the ankle in increased dorsiflexion. 3.2. Comparison with statistical clustering Seventy five of the 1552 limbs (5%) did not match one of the five sagittal gait patterns and were removed from the analysis. Table 2 presents the row and column profiles of the contingency table. The hypothesis of independence between statistical clusters and sagittal gait patterns was rejected (x2 = 1500, p < 0.001). Cluster 3 was related to true equinus, cluster 4 was related to crouch gait and cluster 5 to jump gait. Clusters 1 and 2 were only moderately related to the WNL and apparent equinus gait patterns. Appendix 3, in the supplementary material, presents the correspondence analysis plot of the above contingency table results and the distribution of the clusters on the PFKE scatter plot. 3.3. Kinetics at the knee and sagittal gait patterns Valid kinetic data from 1442 limbs were included. Fig. 2 presents the knee kinetics deviation boxplots for the sagittal gait patterns and normative data. The within normal limit/mild group was the only group not different from normal. Patients in crouch gait had the maximum knee extensor moment. The apparent equinus and jump groups (which were not different from each other) had knee extensor moments on average. However, 40% of the patients in these groups achieved a flexor moment. True equinus was characterized by knee flexor moment which was significantly increased when compared to normal. 3.4. Physical examination and sagittal gait patterns Patients with a complete set of physical examination measurements were included. This resulted in the inclusion of 1366 limbs. Fig. 3 presents the three correspondence analysis plots between physical examination measurements and sagittal gait patterns. The first axes were very dominant, respectively 90%, 90% and 68% of the total variance, and were associated with increasing abnormality of the physical examination measurement. For all plots, the centre of the point cloud was located between the categories 1 (slightly abnormal) and 2 (abnormal). Soleus length discriminated crouch gait from true equinus and jump gait (Fig. 3a). Crouch gait was associated with normal soleus length while jump gait was characterized by soleus contracture. Gastrocsoleus length results were similar (Fig. 3b). Apparent equinus was associated with a low gastrocnemius spasticity score whereas both true equinus and jump gait had high gastrocnemius spasticity scores (Fig. 3c). Crouch gait in Fig. 3c was atypical because of an overrepresentation of a slightly abnormal (1) score without a large component of spasticity overall.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
4
Table 2 Statistical cluster distribution among sagittal gait patterns, column and row profiles of the contingency table. N = 776 patients 2 limbs = 1552. a. Row profiles. Thirty four percent of the overall data were classified as true equinus. Ninety three percent of the data belonging to cluster 3 also belongs to the true equinus sagittal gait pattern. The crouch sagittal gait pattern was related to cluster 4 and the jump gait pattern to cluster 5. b. Column profiles. Thirty one percent of the overall data belong to the first cluster. Sixty four percent of the patient from the within normal limit/mild (WNL) sagittal gait pattern belong to the first cluster. Fifty nine percent of the crouch belongs to cluster 4.
a. Sagial gait paern Crouch Apparent equinus Jump True equinus WNL
----------------- Stascal cluster number ----------------1 2 3 4 5
b. Sagial gait paern Crouch Apparent equinus Jump True equinus WNL Colum average profile
----------------- Stascal cluster number ----------------1 2 3 4
8% 17% 8% 33% 34%
16% 29% 16% 20% 19%
0% 0% 4% 93% 3%
69% 23% 8% 0% 0%
0% 0% 81% 19% 0%
Row average profile 17% 17% 16% 34% 16%
5
16% 32% 15% 30% 64%
26% 48% 28% 16% 32%
0% 0% 5% 49% 3%
59% 20% 7% 0% 0%
0% 0% 46% 5% 0%
31%
27%
18%
14%
9%
4. Discussion This article presented a computational algorithm to classify the sagittal gait patterns of patients with cerebral palsy according to Rodda et al. [2] and the PFKE index was developed to support the algorithm. The PFKE classification algorithm relies solely on the ankle and knee scores which are presented on the PFKE scatter plot. The PFKE index appears effective at classifying the sagittal gait patterns with a 98% accuracy. Patients exhibited a continuum of deviation at the knee and ankle in the middle of stance (Fig. 1 and Appendix 3, in supplementary material). The separation between two groups is
Fig. 2. The average knee moment between 20 and 45% of the gait cycle (N = 776 patients 2 sides = 1552 limbs) and normative data. The graph shows boxplots for the different sagittal gait patterns and normal subjects. The red thick line shows the 95% confidence interval for the normal subjects mean. The letters at the bottom of each boxplot show the results for the Tukey’s statistical group comparison, these are ordered from the largest average knee extensor moment (A, crouch gait) to the smallest (D, true equinus). Boxplots with different letters are significantly different at the a level of 5%. The mild/within normal limit (WNL) group was the only group not different to normal. The group with the maximum knee extensor moment was crouch gait then apparent equinus and jump gait (not different to each other) then WNL. True equinus had the maximum knee flexor moment. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
then somewhat arbitrary and a significant number of patients may be classified in one group while being close to another. The number dPFKE, the 1 D distance to the nearest gait pattern, appraises how close a patient is from the nearest pattern and may be used as a measure of specificity of the gait pattern. It may be applied to select the patients with the most typical gait patterns or to attribute a different weight to data points in statistical analyses. Several studies have devised automatic gait classification systems based on kinematics data, e.g. [12–14]. Such systems present with optimal statistical properties but the statistical process removes the direct correlation with joints function. In most cases, the authors project the various group means back onto the kinematic graphs to assist with clinical understanding and use. Our study followed the opposite approach, it started from the clinically based gait patterns of Rodda et al. and derived the PFKE index to support the automatic classification system. We studied the statistical properties of Rodda’s sagittal gait patterns a-posteriori and compared it to k-means statistical clustering and found notable relationships (Table 2), in particular with respect to the crouch, true equinus and jump gait patterns. However, the PFKE index scatter plot could not discriminate clusters 1 and 2. The introduction of a similar score calculated for the hip allowed to separate these two clusters (Appendix 3, in supplementary material). This is remarkable since the scores were only calculated between 20 and 45% of the gait cycle while clustering takes into account the whole gait cycle. Around 5% of the limbs were located in sections of the PFKE plot that do not correspond to any of the sagittal gait patterns. The limbs presented either (1) within normal limits at the ankle but knee hyper extended or (2) within normal limits at the knee but ankle excessively dorsiflexed. We suggest that limbs in (1) may be called knee recurvatum and that crouch may be extended to limbs in (2). Crouch would then mirror true equinus on the PFKE plot. Knee kinetic results were significantly different according to specific sagittal gait patterns. The three groups with flexed knees (crouch, apparent equinus and jump) had increased knee extensor moment with respect to normal. However, most patients with crouch gait did not achieve a knee flexor moment during stance whereas about 40% of the patients in apparent equinus or jump gait were able to cross over from knee extensor to knee flexor moment. This highlights the role of the plantarflexors acting across the ankle
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
5
Fig. 3. Correspondence analysis plots (‘french’ symmetrical mode) between physical examination measurements (soleus length, gastrocsoleus length and gastrocnemius spasticity) and sagittal gait patterns. The plots allow the visualization of the cross-tabulation (physical examination sagittal gait pattern) point clouds in the plane that maximize the variance. Red triangles display grades of abnormal values for the physical examination measurements (0, normal to 4, extremely abnormal) while blue dots represent sagittal gait patterns. The variance explained by each axis is provided in parentheses. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
joint to direct the position of the ground reaction force with respect to the knee joint. In 1974 Perry described the importance of the ankle second rocker [22], and in 1991 Gage emphasized the role of the plantarflexors during second rocker [23]. Gage described the plantarflexion–knee extension couple, a mechanism whereby the gastrocnemius–soleus complex controls the advancement of the tibia over the foot, the extension of the knee and the orientation of the ground reaction vector with respect to the knee. We entirely agree with the description of the mechanism but found the terminology confusing. During second rocker there is no plantarflexion but dorsiflexion at the ankle and there is either a dorsiflexing moment of the ground reaction force or a plantarflexors moment at the ankle. We decided to keep the reference to the plantarflexors and named the index ‘‘plantarflexor–knee extension couple’’. We chose the period 20–45% of the gait cycle for the calculations to avoid some of the kinematic and kinetic features that are consequences of the loading response, such as the peak knee flexion and knee extensor moment in stance (Appendix 1, in supplementary material). We also observed that the ankle dorsiflexion curve showed a different rate of change between the 10–20% and 20–45% periods. This may be related to the kinematics of the foot in the sagittal plane, where the foot lies flat on the ground only between 10 and 20% before it rises slowly between 20 and 45%. It may be that the gastrocnemius–soleus
complex only actively controls the advancement of the tibia over the foot between 20 and 45% rather than 10–45%. Given the major role attributed to the plantarflexors we studied the link between physical examination measurements and sagittal gait patterns. A five point ordinal scale was fitted to the raw physical examination measurements (Table 1) to standardize statistical treatment of the data. Results showed that the jump gait pattern was associated with soleus and gastrocnemius contracture as well as gastrocnemius spasticity. True equinus was correlated with moderate soleus and gastrocnemius contracture and gastrocnemius spasticity. Spasticity may be the main driver of the true equinus gait pattern. Unsurprisingly, patients in crouch gait had no contracture of the soleus or gastrocnemius muscles. Patients in crouch gait were also overrepresented in the slightly abnormal spasticity group (score 1, Fig. 3c) but their overall spasticity measures were only mildly abnormal. Patients in crouch gait may have occult spasticity which may not be triggered due to their functionally long gastrocnemius and soleus muscles. The apparent equinus group presented only slightly abnormal measurements for all three measurements. We conclude the muscle properties of the plantarflexors may not be the main cause for the deviations found in the knee kinematics. We suspect most patients in apparent equinus walked with flexed knees because of hamstring spasticity or hamstring contracture. This has major implications in terms of management algorithms.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
6
5. Conclusion The plantarflexor–knee extension couple index presented in this study allows computational classification of sagittal gait patterns in cerebral palsy. This may be useful in the stratification of cohorts of patients with cerebral palsy in clinical trials. The clinically derived sagittal gait patterns were correlated with statistical classification. Correspondence analysis between physical examination measurements and sagittal gait patterns highlighted the determinant role played by the gastrocnemius and soleus muscles except for patients exhibiting the apparent equinus gait pattern. Acknowledgments This work has only been possible with the support of the staff of the Hugh Williamson Gait Analysis Laboratory. Their assistance with data collection is gratefully acknowledged. This study was partly funded from a grant from the Clinical Science theme, the Murdoch Childrens Research Institute. Conflict of interest statement The authors declare no conflict of interest to disclose.
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.gaitpost.2014.12. 019. References [1] Davids JR, Ounpuu S, DeLuca PA, Davis III RB. Optimization of walking ability of children with cerebral palsy. Instr Course Lect 2004;53:511–22. [2] Rodda JM, Graham HK, Carson L, Galea MP, Wolfe R. Sagittal gait patterns in spastic diplegia. J Bone Joint Surg Br 2004;86(2):251–8. [3] Rodda J, Graham HK. Classification of gait patterns in spastic hemiplegia and spastic diplegia: a basis for a management algorithm. Eur J Neurol 2001;8(Suppl. 5):98–108.
[4] Rang M. Cerebral palsy. In: Lovell WW, Winter RB, editors. Pediatric orthopaedics. 3rd ed., Philadelphia: J.B. Lippincott; 1990. p. 465–506. [5] Sutherland DH, Davids JR. Common gait abnormalities of the knee in cerebral palsy. Clin Orthop 1993;288:139–47. [6] Miller F, Dabney KW, Rang M. Complications in cerebral palsy treatment. In: Epps CH, Bowen JR, editors. Complications in pediatric orthopaedic surgery. Philadelphia: J.B. Lippincott Company; 1995. p. 477–544. [7] Thomason P, Graham HK. In: Iansek R, Morris ME, editors. Rehabilitation of children with cerebral palsy after single-event multilevel surgery. 2013. [8] Rutz E, Tirosh O, Thomason P, Barg A, Graham HK. Stability of the gross motor function classification system after single-event multilevel surgery in children with cerebral palsy. Dev Med Child Neurol 2012;54(12):1109–13. [9] Opheim A, McGinley JL, Olsson E, Stanghelle JK, Jahnsen R. Walking deterioration and gait analysis in adults with spastic bilateral cerebral palsy. Gait Posture 2013;37(2):165–71. [10] Kadhim M, Miller F. Crouch gait changes after planovalgus foot deformity correction in ambulatory children with cerebral palsy. Gait Posture 2014;39(2):793–8. [11] Franki I, De Cat J, Deschepper E, Molenaers G, Desloovere K, Himpens E, et al. A clinical decision framework for the identification of main problems and treatment goals for ambulant children with bilateral spastic cerebral palsy. Res Dev Disabil 2014;35(5):1160–76. [12] O’Byrne JM, Jenkinson A, O’Brien TM. Quantitative analysis and classification of gait patterns in cerebral palsy using a three-dimensional motion analyzer. J Child Neurol 1998;13(3):101–8. [13] Rozumalski A, Schwartz MH. Crouch gait patterns defined using k-means cluster analysis are related to underlying clinical pathology. Gait Posture 2009;30(2):155–60. [14] Bonnefoy-Mazure A, Sagawa Jr Y, Lascombes P, De Coulon G, Armand S. Identification of gait patterns in individuals with cerebral palsy using multiple correspondence analysis. Res Dev Disabil 2013;34(9):2684–93. [15] Gage JR. The clinical use of kinetics for evaluation of pathologic gait in cerebral palsy. Instr Course Lect 1995;44:507–15. [16] Schwartz MH, Rozumalski A. The gait deviation index: a new comprehensive index of gait pathology. Gait Posture 2008;28(3):351–7. [17] R Core Team.. R: a language and environment for statistical computing. R Foundation for Statistical Computing; 2014. [18] Keenan WN, Rodda J, Wolfe R, Roberts S, Borton DC, Graham HK. The static examination of children and young adults with cerebral palsy in the gait analysis laboratory: technique and observer agreement. J Pediatr Orthop B 2004;13:1–8. [19] Tardieu G, Shentoub S, Delarue R. a La Recherche Dune Technique De Mesure De La Spasticite. Rev Neurol 1954;91(2):143–4. [20] Mudge AJ, Bau KV, Purcell LN, Wu JC, Axt MW, Selber P, et al. Normative reference values for lower limb joint range, bone torsion, and alignment in children aged 4–16 years. J Pediatr Orthop B 2014;23(1):15–25. [21] Leˆ S, Josse J, Husson F. FactoMineR: an R package for multivariate analysis. J Stat Softw 2008;25(1):1–18. [22] Perry J. Kinesology of lower-extremity bracing. Clin Orthop Relat Res 1974;102:18–31. [23] Gage JR. Gait analysis in cerebral palsy. London: Mac Keith Press; 1991.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
GAIPOS-4384; No. of Pages 6 Gait & Posture xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Gait & Posture journal homepage: www.elsevier.com/locate/gaitpost
Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index Morgan Sangeux a,b,c,*, Jill Rodda a,b,c, H. Kerr Graham a,b,c a
The Royal Children’s Hospital, Australia Murdoch Childrens Research Institute, Australia c The University of Melbourne, Australia b
A R T I C L E I N F O
A B S T R A C T
Article history: Received 1 September 2014 Received in revised form 23 December 2014 Accepted 30 December 2014
The identification of gait patterns in cerebral palsy offers a common language for clinicians and contributes to management algorithms. We describe a quantitative classification of sagittal gait patterns based on the plantarflexor–knee extension couple index. This consists of a scatter plot based on ankle and knee scores, and allows objective identification of the sagittal gait pattern. Sagittal kinematic data from 200 limbs of 100 patients with bilateral spastic cerebral palsy were utilized to validate the algorithm against the assessment of a clinician with expertise in gait pattern identification. A dataset of 776 cerebral palsy patients, 1552 limbs, was used to compare the sagittal gait patterns against k-means statistical clustering. The classification was further explored with respect to the knee kinetics during the middle of stance and physical examination measurements of the gastrocnemius–soleus complex. Two supplementary materials (Appendices 2 and 3) provide in-depth discussion about statistical properties of the plantarflexor–knee extension couple index as well as its relationship with statistical clustering. The plantarflexor–knee extension index achieved 98% accuracy and may be suitable for the computational classification of large patient cohorts and multicentre studies. The sagittal gait patterns were strongly related to k-means statistical clustering and physical examination of the gastrocnemius– soleus complex. Patients in crouch gait had normal soleus and gastrocnemius lengths but spasticity in the gastrocnemius. Patients in jump gait exhibited a short gastrocnemius and soleus and gastrocnemius spasticity. Patients in true equinus presented with a moderately contracted soleus and gastrocnemius and gastrocnemius spasticity. Patients in apparent equinus did not show abnormal physical examination measurements for the gastrocnemius–soleus complex. ß 2015 Elsevier B.V. All rights reserved.
Keywords: Gait Pattern Index Physical examination Cerebral palsy
1. Introduction Davids et al. [1] described five sources of data to guide clinical decision-making for children with cerebral palsy. One of these, instrumented gait analysis, provides detailed information on the kinematics and kinetics of the joints of the lower limb. A typical instrumented gait analysis entails, for each limb, to analyze at least ten kinematics and kinetics curves. The interpretation of this data requires linking kinematic deviations with physical examination measurements, to define the gait impairments. The amount of data
* Corresponding author at: The Hugh Williamson Gait Analysis Laboratory, The Royal Children’s Hospital, 50 Flemington Road, Parkville, VIC 3052, Australia. Tel.: +61 416910735. E-mail address: [email protected] (M. Sangeux).
is invaluable to determine the appropriate treatment for a specific patient but may make it difficult to identify patterns and the construction of management algorithms. Rodda et al. [2,3], described a semi-quantitative sagittal gait pattern classification for patients with bilateral spastic cerebral palsy (BSCP). That classification combined pattern recognition with quantitative kinematic data, and was based on an extension of earlier work by Rang et al. [4], Sutherland and Davids [5] and Miller et al. [6]. The Rodda classification described five groups: crouch gait, jump gait, apparent equinus, true equinus and mild gait (within normal limits in the sagittal plane). The classification in five groups applies to the limb and the asymmetric group is introduced when the two limbs of the same patient belong to two different classifications. Based on the above sagittal gait patterns Rodda et al. derived a management algorithm which specifies the dominant muscle groups to be targeted for treatment of spasticity
http://dx.doi.org/10.1016/j.gaitpost.2014.12.019 0966-6362/ß 2015 Elsevier B.V. All rights reserved.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
2
or contracture and includes prescription of orthotics. This work has often been utilized to characterize the gait patterns of cohorts of patients (e.g. [7–11]). Although the classification was based in part on quantitative data, it is also in part qualitative and subjective. It therefore requires the involvement of a clinician who is expert in gait analysis and in clinical assessment. This requirement may restrict the use of the Rodda classification to expert clinicians and may prohibit its application in large patient cohorts. Therefore the first purpose of this study was to propose and validate an algorithm to classify Rodda’s sagittal gait patterns on a purely quantitative basis. Quantitative algorithms based on clustering of the kinematic data have been proposed previously (e.g. [12–14]). These algorithms present optimal statistical properties but may be difficult to apply clinically and they have seldom been used in management algorithms. The sagittal gait classification by Rodda was not derived from statistical clustering but from years of clinical observations and the underlying statistical properties are unknown. Therefore the second purpose of this study was to compare the Rodda sagittal gait pattern classification with statistical clustering. Sagittal plane kinematics at the ankle and knee in stance are mainly determined by the plantarflexion–knee extension couple [15]. The couple refers to the action of the gastrocnemius and soleus muscles, the ankle plantarflexors, to control both the advancement of the tibia over the foot and the knee kinetics in mid-stance. Spasticity or contracture of the gastrocnemius–soleus muscles in children with cerebral palsy is frequently present and will influence sagittal kinematics and kinetics. The third purpose of this study was to compare physical examination measurements of the plantarflexors with the sagittal gait patterns. 2. Material and methods 2.1. The plantarflexor–knee extension couple index The plantarflexor–knee extension (PFKE) index calculates the distance of the patient’s ankle and knee kinematics in mid-stance
from normative data. The period of the gait cycle used to calculate the PFKE index is set between 20 and 45% of the gait cycle (see Appendix 1, in supplementary material, for a discussion about this choice). During this period, the knee extends while the ankle dorsiflexes, the knee moment changes from an extensor moment to a flexor moment allowing the quadriceps to cease contracting and the ankle to absorb power through the eccentric contraction of the gastrocnemius–soleus complex. The PFKE index consists of two scores from the ankle and the knee. It is calculated as follow: 45 X kci mci 1 PFKEc ¼ 45 20 þ 1 i¼20 s ci where superscript c denotes the knee or ankle curve, subscript i denotes the time instant (in % of the gait cycle), ki is the value of the kinematic curve at the i% instant of the gait cycle, mi is the value of the normal kinematic curve at the i% instant and si is the value of the standard deviation from the normal kinematics curve at the i% instant. A PFKE index of (3,2) means that, on average between 20 and 45% of the gait cycle, the ankle curve is three standard deviation below normal and the knee curve is two standard deviations above normal. The normal kinematics curves originated from 35 typically developed children (17 girls, 18 boys) with an age range of 6–17 years old. Classification is based on the +1 or 1 values of the PFKE ankle and knee scores and is displayed on the PFKE scatter plot (Fig. 1 and Appendix 2 in the supplementary material). We define dPFKE as the minimum 1 D distance of one point to the other gait patterns. A large value for dPFKE means the point is located far from all the other gait classifications and therefore specific to the gait pattern it belongs to. Validation of the classification algorithm was performed by comparing the classification by the algorithm to classification by a clinical expert (J.R.) in a study of 200 lower limbs, in 100 patients with BSCP. Kinematic data from patients diagnosed with BSCP were randomly selected from our database. Accuracy of the classifier was expressed as the percentage of true classifications divided by the number of limbs assessed.
Fig. 1. The plantarflexor–knee extension scatter plot, with the ankle score in X and knee score in Y, for N = 200 limbs. The lines, corresponding to the 1 or 1 value for the ankle and knee scores, separate the sagittal gait patterns. The central section of the PFKE plot corresponds to the within normal limit (WNL) or mild sagittal gait pattern. An example for a PFKE index of (3,2) is provided. This point corresponds to the jump gait pattern and is at a minimum distance dPFKE = 1 to the nearest gait pattern (true equinus). The validation dataset is overlayed on the plot. Points were coloured in green if the automatic and expert classifications agreed and red if they disagreed. It can be noticed than only 4 points were misclassified and those were very close to the line between two classifications. This is confirmed by an average dPFKE for those 4 points of 0.2. Eight points were located in areas of the plot without a labelled gait pattern. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
2.2. Relationship with statistical classification A larger dataset was used for the remaining of this study. All patients with a diagnosis of cerebral palsy for which kinematics and kinetics gait data captured barefoot between January 2004 and December 2013 were included. Previous surgery, including plantarflexor lengthening was not an exclusion criterion for this study. Seven hundred and seventy six patients, 1552 limbs, were included. k-means clustering was chosen as the statistical classification algorithm and the methods described by Rozumalski et al. [13] were followed. Readers are referred to the original publication for specific details. Briefly, one column vector was constructed from the pelvis, hip, knee and ankle sagittal kinematics of each limb. Kinematic data were sampled for each percent of the gait cycle which lead to a vector of 4 curves 101 points = 404 points. Column vectors from the 1552 limbs were adjoined to form a 404 1552 matrix, KD. A reduced order approximation was obtained by projecting KD on the four orthogonal axes (eigenvectors or gait features [13,16]) which accounted for 96% of the total variance. k-means clustering requires predefining the number of clusters. Since the sagittal gait pattern classification has five groups, the number k of clusters was set to five. Calculations were performed in the statistical software R version 3.0.3 [17]. To compare the two classifications, a contingency table (or cross tabulation) was utilized. The contingency table C has five rows, one for each sagittal gait pattern, and five columns, one for each k-means clusters. Each element, Cij, of C corresponds to the number of limbs belonging to both the sagittal gait pattern i and the cluster j. Pearson’s x2 test, applied to C, tested for the independence between the two classifications. 2.3. Knee kinetic and physical examination of the gastrocnemius and soleus muscles The average sagittal knee moment between 20 and 45% of the gait cycle was calculated and compared between sagittal gait patterns and normative knee kinetic data. One way ANOVA and Tukey’s post-hoc group comparison was performed in Minitab (Minitab Inc., USA). Physical examination measurements relating to the lengths of the gastrocnemius and soleus muscles as well as gastrocnemius spasticity were extracted from the database. Soleus length was defined as the maximal ankle dorsiflexion angle, subject supine with the knee flexed at 908 [18]. Gastrocsoleus length was expressed as the maximal ankle dorsiflexion, subject supine with the knee extended (Angle_KneeExt [18]). Gastrocnemius spasticity was assessed using the Tardieu test, with the patient supine and the knee extended [19]. The ankle was dorsiflexed rapidly to elicit a ‘‘catch’’ which was then measured using a goniometer (Angle_Tardieu). To grade spasticity, we used the arithmetic difference Angle_KneeExt – Angle_Tardieu. An ordinal scale was fitted to the raw data with five categories: normal (0), slightly abnormal (1), abnormal (2), very abnormal (4) and extremely abnormal (5). Table 1 presents the angle ranges that define the scale. These values were derived from [20] and clinical experience. We utilized correspondence analysis, a form of correlation analysis when data are nominal, to study the association between physical examination measurements and sagittal gait patterns. Correspondence analysis allows the best possible graphical representation of contingency tables in a plane. All calculations were performed in R with the package FactoMineR [21].
3
Table 1 Relationship between the scale (ordinal nominal) and physical examination measurements angles at the ankle. The first column presents values for the soleus measurement (knee flexed at 908) and the second column values for the gastrocsoleus measurement (knee extended). The third column presents values for the difference between the gastrocsoleus measurement (static, knee extended) and the Tardieu measurement (dynamic, knee extended). Subjects are considered within normal limits (scale = 0) for: a soleus measurement greater than 258 dorsiflexion, a gastrocsoleus measurement greater than 158 dorsiflexion and a difference between the static and the dynamic measurements less than 58. Description
Normal Slightly abnormal Abnormal Very abnormal Extremely abnormal
Scale
Physical examination measurements Ankle angle (8) knee flexed at 908
Ankle angle (8) knee extended
Ankle angle difference (8) static–dynamic
0 1
25 < x 10 < x < 25
15 < x 0 < x < 15
x<5 5 < x < 10
2 3 4
0 < x < 10 10 < x < 0 Less than 10
10 < x < 0 20 < x < 10 Less than 20
10 < x < 20 20 < x < 30 30 and over
3. Results 3.1. Accuracy of the automatic classification based on the PFKE index Fig. 1 presents the four false classifications leading to 4/200 = 98% accuracy. Patients presented a continuum of deviation from normal at the knee and ankle rather than well delineated groups in separated portions of the graph. The average dPFKE for misclassified patients was 0.2. The PFKE index for eight limbs were located in sections of the plot that do not correspond to any previous sagittal gait pattern classification. These correspond to the ankle within normal limits and the knee in recurvatum or the knee within normal limits and the ankle in increased dorsiflexion. 3.2. Comparison with statistical clustering Seventy five of the 1552 limbs (5%) did not match one of the five sagittal gait patterns and were removed from the analysis. Table 2 presents the row and column profiles of the contingency table. The hypothesis of independence between statistical clusters and sagittal gait patterns was rejected (x2 = 1500, p < 0.001). Cluster 3 was related to true equinus, cluster 4 was related to crouch gait and cluster 5 to jump gait. Clusters 1 and 2 were only moderately related to the WNL and apparent equinus gait patterns. Appendix 3, in the supplementary material, presents the correspondence analysis plot of the above contingency table results and the distribution of the clusters on the PFKE scatter plot. 3.3. Kinetics at the knee and sagittal gait patterns Valid kinetic data from 1442 limbs were included. Fig. 2 presents the knee kinetics deviation boxplots for the sagittal gait patterns and normative data. The within normal limit/mild group was the only group not different from normal. Patients in crouch gait had the maximum knee extensor moment. The apparent equinus and jump groups (which were not different from each other) had knee extensor moments on average. However, 40% of the patients in these groups achieved a flexor moment. True equinus was characterized by knee flexor moment which was significantly increased when compared to normal. 3.4. Physical examination and sagittal gait patterns Patients with a complete set of physical examination measurements were included. This resulted in the inclusion of 1366 limbs. Fig. 3 presents the three correspondence analysis plots between physical examination measurements and sagittal gait patterns. The first axes were very dominant, respectively 90%, 90% and 68% of the total variance, and were associated with increasing abnormality of the physical examination measurement. For all plots, the centre of the point cloud was located between the categories 1 (slightly abnormal) and 2 (abnormal). Soleus length discriminated crouch gait from true equinus and jump gait (Fig. 3a). Crouch gait was associated with normal soleus length while jump gait was characterized by soleus contracture. Gastrocsoleus length results were similar (Fig. 3b). Apparent equinus was associated with a low gastrocnemius spasticity score whereas both true equinus and jump gait had high gastrocnemius spasticity scores (Fig. 3c). Crouch gait in Fig. 3c was atypical because of an overrepresentation of a slightly abnormal (1) score without a large component of spasticity overall.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
4
Table 2 Statistical cluster distribution among sagittal gait patterns, column and row profiles of the contingency table. N = 776 patients 2 limbs = 1552. a. Row profiles. Thirty four percent of the overall data were classified as true equinus. Ninety three percent of the data belonging to cluster 3 also belongs to the true equinus sagittal gait pattern. The crouch sagittal gait pattern was related to cluster 4 and the jump gait pattern to cluster 5. b. Column profiles. Thirty one percent of the overall data belong to the first cluster. Sixty four percent of the patient from the within normal limit/mild (WNL) sagittal gait pattern belong to the first cluster. Fifty nine percent of the crouch belongs to cluster 4.
a. Sagial gait paern Crouch Apparent equinus Jump True equinus WNL
----------------- Stascal cluster number ----------------1 2 3 4 5
b. Sagial gait paern Crouch Apparent equinus Jump True equinus WNL Colum average profile
----------------- Stascal cluster number ----------------1 2 3 4
8% 17% 8% 33% 34%
16% 29% 16% 20% 19%
0% 0% 4% 93% 3%
69% 23% 8% 0% 0%
0% 0% 81% 19% 0%
Row average profile 17% 17% 16% 34% 16%
5
16% 32% 15% 30% 64%
26% 48% 28% 16% 32%
0% 0% 5% 49% 3%
59% 20% 7% 0% 0%
0% 0% 46% 5% 0%
31%
27%
18%
14%
9%
4. Discussion This article presented a computational algorithm to classify the sagittal gait patterns of patients with cerebral palsy according to Rodda et al. [2] and the PFKE index was developed to support the algorithm. The PFKE classification algorithm relies solely on the ankle and knee scores which are presented on the PFKE scatter plot. The PFKE index appears effective at classifying the sagittal gait patterns with a 98% accuracy. Patients exhibited a continuum of deviation at the knee and ankle in the middle of stance (Fig. 1 and Appendix 3, in supplementary material). The separation between two groups is
Fig. 2. The average knee moment between 20 and 45% of the gait cycle (N = 776 patients 2 sides = 1552 limbs) and normative data. The graph shows boxplots for the different sagittal gait patterns and normal subjects. The red thick line shows the 95% confidence interval for the normal subjects mean. The letters at the bottom of each boxplot show the results for the Tukey’s statistical group comparison, these are ordered from the largest average knee extensor moment (A, crouch gait) to the smallest (D, true equinus). Boxplots with different letters are significantly different at the a level of 5%. The mild/within normal limit (WNL) group was the only group not different to normal. The group with the maximum knee extensor moment was crouch gait then apparent equinus and jump gait (not different to each other) then WNL. True equinus had the maximum knee flexor moment. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
then somewhat arbitrary and a significant number of patients may be classified in one group while being close to another. The number dPFKE, the 1 D distance to the nearest gait pattern, appraises how close a patient is from the nearest pattern and may be used as a measure of specificity of the gait pattern. It may be applied to select the patients with the most typical gait patterns or to attribute a different weight to data points in statistical analyses. Several studies have devised automatic gait classification systems based on kinematics data, e.g. [12–14]. Such systems present with optimal statistical properties but the statistical process removes the direct correlation with joints function. In most cases, the authors project the various group means back onto the kinematic graphs to assist with clinical understanding and use. Our study followed the opposite approach, it started from the clinically based gait patterns of Rodda et al. and derived the PFKE index to support the automatic classification system. We studied the statistical properties of Rodda’s sagittal gait patterns a-posteriori and compared it to k-means statistical clustering and found notable relationships (Table 2), in particular with respect to the crouch, true equinus and jump gait patterns. However, the PFKE index scatter plot could not discriminate clusters 1 and 2. The introduction of a similar score calculated for the hip allowed to separate these two clusters (Appendix 3, in supplementary material). This is remarkable since the scores were only calculated between 20 and 45% of the gait cycle while clustering takes into account the whole gait cycle. Around 5% of the limbs were located in sections of the PFKE plot that do not correspond to any of the sagittal gait patterns. The limbs presented either (1) within normal limits at the ankle but knee hyper extended or (2) within normal limits at the knee but ankle excessively dorsiflexed. We suggest that limbs in (1) may be called knee recurvatum and that crouch may be extended to limbs in (2). Crouch would then mirror true equinus on the PFKE plot. Knee kinetic results were significantly different according to specific sagittal gait patterns. The three groups with flexed knees (crouch, apparent equinus and jump) had increased knee extensor moment with respect to normal. However, most patients with crouch gait did not achieve a knee flexor moment during stance whereas about 40% of the patients in apparent equinus or jump gait were able to cross over from knee extensor to knee flexor moment. This highlights the role of the plantarflexors acting across the ankle
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
5
Fig. 3. Correspondence analysis plots (‘french’ symmetrical mode) between physical examination measurements (soleus length, gastrocsoleus length and gastrocnemius spasticity) and sagittal gait patterns. The plots allow the visualization of the cross-tabulation (physical examination sagittal gait pattern) point clouds in the plane that maximize the variance. Red triangles display grades of abnormal values for the physical examination measurements (0, normal to 4, extremely abnormal) while blue dots represent sagittal gait patterns. The variance explained by each axis is provided in parentheses. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
joint to direct the position of the ground reaction force with respect to the knee joint. In 1974 Perry described the importance of the ankle second rocker [22], and in 1991 Gage emphasized the role of the plantarflexors during second rocker [23]. Gage described the plantarflexion–knee extension couple, a mechanism whereby the gastrocnemius–soleus complex controls the advancement of the tibia over the foot, the extension of the knee and the orientation of the ground reaction vector with respect to the knee. We entirely agree with the description of the mechanism but found the terminology confusing. During second rocker there is no plantarflexion but dorsiflexion at the ankle and there is either a dorsiflexing moment of the ground reaction force or a plantarflexors moment at the ankle. We decided to keep the reference to the plantarflexors and named the index ‘‘plantarflexor–knee extension couple’’. We chose the period 20–45% of the gait cycle for the calculations to avoid some of the kinematic and kinetic features that are consequences of the loading response, such as the peak knee flexion and knee extensor moment in stance (Appendix 1, in supplementary material). We also observed that the ankle dorsiflexion curve showed a different rate of change between the 10–20% and 20–45% periods. This may be related to the kinematics of the foot in the sagittal plane, where the foot lies flat on the ground only between 10 and 20% before it rises slowly between 20 and 45%. It may be that the gastrocnemius–soleus
complex only actively controls the advancement of the tibia over the foot between 20 and 45% rather than 10–45%. Given the major role attributed to the plantarflexors we studied the link between physical examination measurements and sagittal gait patterns. A five point ordinal scale was fitted to the raw physical examination measurements (Table 1) to standardize statistical treatment of the data. Results showed that the jump gait pattern was associated with soleus and gastrocnemius contracture as well as gastrocnemius spasticity. True equinus was correlated with moderate soleus and gastrocnemius contracture and gastrocnemius spasticity. Spasticity may be the main driver of the true equinus gait pattern. Unsurprisingly, patients in crouch gait had no contracture of the soleus or gastrocnemius muscles. Patients in crouch gait were also overrepresented in the slightly abnormal spasticity group (score 1, Fig. 3c) but their overall spasticity measures were only mildly abnormal. Patients in crouch gait may have occult spasticity which may not be triggered due to their functionally long gastrocnemius and soleus muscles. The apparent equinus group presented only slightly abnormal measurements for all three measurements. We conclude the muscle properties of the plantarflexors may not be the main cause for the deviations found in the knee kinematics. We suspect most patients in apparent equinus walked with flexed knees because of hamstring spasticity or hamstring contracture. This has major implications in terms of management algorithms.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019
G Model
GAIPOS-4384; No. of Pages 6 M. Sangeux et al. / Gait & Posture xxx (2015) xxx–xxx
6
5. Conclusion The plantarflexor–knee extension couple index presented in this study allows computational classification of sagittal gait patterns in cerebral palsy. This may be useful in the stratification of cohorts of patients with cerebral palsy in clinical trials. The clinically derived sagittal gait patterns were correlated with statistical classification. Correspondence analysis between physical examination measurements and sagittal gait patterns highlighted the determinant role played by the gastrocnemius and soleus muscles except for patients exhibiting the apparent equinus gait pattern. Acknowledgments This work has only been possible with the support of the staff of the Hugh Williamson Gait Analysis Laboratory. Their assistance with data collection is gratefully acknowledged. This study was partly funded from a grant from the Clinical Science theme, the Murdoch Childrens Research Institute. Conflict of interest statement The authors declare no conflict of interest to disclose.
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.gaitpost.2014.12. 019. References [1] Davids JR, Ounpuu S, DeLuca PA, Davis III RB. Optimization of walking ability of children with cerebral palsy. Instr Course Lect 2004;53:511–22. [2] Rodda JM, Graham HK, Carson L, Galea MP, Wolfe R. Sagittal gait patterns in spastic diplegia. J Bone Joint Surg Br 2004;86(2):251–8. [3] Rodda J, Graham HK. Classification of gait patterns in spastic hemiplegia and spastic diplegia: a basis for a management algorithm. Eur J Neurol 2001;8(Suppl. 5):98–108.
[4] Rang M. Cerebral palsy. In: Lovell WW, Winter RB, editors. Pediatric orthopaedics. 3rd ed., Philadelphia: J.B. Lippincott; 1990. p. 465–506. [5] Sutherland DH, Davids JR. Common gait abnormalities of the knee in cerebral palsy. Clin Orthop 1993;288:139–47. [6] Miller F, Dabney KW, Rang M. Complications in cerebral palsy treatment. In: Epps CH, Bowen JR, editors. Complications in pediatric orthopaedic surgery. Philadelphia: J.B. Lippincott Company; 1995. p. 477–544. [7] Thomason P, Graham HK. In: Iansek R, Morris ME, editors. Rehabilitation of children with cerebral palsy after single-event multilevel surgery. 2013. [8] Rutz E, Tirosh O, Thomason P, Barg A, Graham HK. Stability of the gross motor function classification system after single-event multilevel surgery in children with cerebral palsy. Dev Med Child Neurol 2012;54(12):1109–13. [9] Opheim A, McGinley JL, Olsson E, Stanghelle JK, Jahnsen R. Walking deterioration and gait analysis in adults with spastic bilateral cerebral palsy. Gait Posture 2013;37(2):165–71. [10] Kadhim M, Miller F. Crouch gait changes after planovalgus foot deformity correction in ambulatory children with cerebral palsy. Gait Posture 2014;39(2):793–8. [11] Franki I, De Cat J, Deschepper E, Molenaers G, Desloovere K, Himpens E, et al. A clinical decision framework for the identification of main problems and treatment goals for ambulant children with bilateral spastic cerebral palsy. Res Dev Disabil 2014;35(5):1160–76. [12] O’Byrne JM, Jenkinson A, O’Brien TM. Quantitative analysis and classification of gait patterns in cerebral palsy using a three-dimensional motion analyzer. J Child Neurol 1998;13(3):101–8. [13] Rozumalski A, Schwartz MH. Crouch gait patterns defined using k-means cluster analysis are related to underlying clinical pathology. Gait Posture 2009;30(2):155–60. [14] Bonnefoy-Mazure A, Sagawa Jr Y, Lascombes P, De Coulon G, Armand S. Identification of gait patterns in individuals with cerebral palsy using multiple correspondence analysis. Res Dev Disabil 2013;34(9):2684–93. [15] Gage JR. The clinical use of kinetics for evaluation of pathologic gait in cerebral palsy. Instr Course Lect 1995;44:507–15. [16] Schwartz MH, Rozumalski A. The gait deviation index: a new comprehensive index of gait pathology. Gait Posture 2008;28(3):351–7. [17] R Core Team.. R: a language and environment for statistical computing. R Foundation for Statistical Computing; 2014. [18] Keenan WN, Rodda J, Wolfe R, Roberts S, Borton DC, Graham HK. The static examination of children and young adults with cerebral palsy in the gait analysis laboratory: technique and observer agreement. J Pediatr Orthop B 2004;13:1–8. [19] Tardieu G, Shentoub S, Delarue R. a La Recherche Dune Technique De Mesure De La Spasticite. Rev Neurol 1954;91(2):143–4. [20] Mudge AJ, Bau KV, Purcell LN, Wu JC, Axt MW, Selber P, et al. Normative reference values for lower limb joint range, bone torsion, and alignment in children aged 4–16 years. J Pediatr Orthop B 2014;23(1):15–25. [21] Leˆ S, Josse J, Husson F. FactoMineR: an R package for multivariate analysis. J Stat Softw 2008;25(1):1–18. [22] Perry J. Kinesology of lower-extremity bracing. Clin Orthop Relat Res 1974;102:18–31. [23] Gage JR. Gait analysis in cerebral palsy. London: Mac Keith Press; 1991.
Please cite this article in press as: Sangeux M, et al. Sagittal gait patterns in cerebral palsy: The plantarflexor–knee extension couple index. Gait Posture (2015), http://dx.doi.org/10.1016/j.gaitpost.2014.12.019