What’s Known

The movement system impairment model has certain limitations in the classification of knee impairments. Partitioning around medoids clustering is a method to organize homogeneous groups with similar characteristics.

What’s New

A new clinical knee examination model is introduced. The model uses the partitioning around medoids clustering method based on clinical signs and symptoms. The model successfully reduced the number of clinical examination variables and prioritized discriminative variables.

Introduction

Extra-articular soft tissue injuries are the most common injuries of the knee joint, which may eventually lead to knee osteoarthritis with an estimated prevalence of 24% in adults. ^{1 , 2} Knee pain is a common musculoskeletal problem in patients referred to outpatient physical therapy centers. The prevalence of knee pain in the United States has increased by 65% over 20 years. ³ Conservative treatment of non-traumatic knee pain is usually the first line of management versus surgical treatments. ⁴ Over the years, various non-surgical treatments with an emphasis on correcting the impaired movement have been developed for knee pain. ⁵ Despite the availability of various treatments, it has been suggested that classifying patients into homogeneous groups may simplify the complexities of diagnosis and facilitate treatment options. ^{5 - 7}

Clinical methods classify patients based on their distinct signs, symptoms, treatments, or psychosocial characteristics. ^{8 , 9} Among the most important clinical classification methods are the pathoanatomic and kinesiopathologic models. ^{10 - 12} The pathoanatomic model focuses on the effect of stresses on tissues and utilizes diagnostic labeling of the causative factors of the patient’s pain (e.g., meniscal tear). However, sometimes patients are referred to a physical therapist with no definite diagnosis. Physical therapists typically treat patients based on their movement impairments rather than on the pathoanatomical source of the pain. As a result, they use the kinesiopathologic model, a movement system diagnostic classification method, as an alternative to the pathoanatomic model. ¹¹

Among kinesiopathologic models, the movement system impairment (MSI) model is considered the most suitable classification approach and has been reviewed clinically and kinematically during the current decade. ^{5 , 13 , 14} The MSI model of the knee has been shown to have acceptable intratester reliability. ¹⁴ This model classifies patients with knee pain into seven subgroups based on examining the knee movement, correction of faulty posture, and re-evaluation of the symptoms. The subgroups are tibiofemoral rotation syndrome, tibiofemoral hypomobility syndrome, tibiofemoral accessory hypermobility syndrome, knee extension syndrome; knee hyperextension syndrome, patellar lateral glide syndrome, and knee impairment. Previous studies have reported certain limitations in the MSI model, namely overlapping signs and symptoms, ¹³ time-consuming examination process, ⁵ challenges in determining the tibiofemoral rotation angle, ⁴ and the absence of differences in perceived pain levels between the usual and corrected movement conditions. ¹³

Attempts have been made to validate the classification of syndromes using statistical models such as Ward’s clustering method for hip disorders ¹⁵ and hierarchical clustering method for non-specific patellofemoral pain for the lower extremity examination. ¹⁶ Statistical analysis can be used to describe the characteristics of a population in an idealized model with algorithms based on similarity and differences in signs and symptoms. ^{17 , 18} Statistical clustering is an unsupervised learning process in which the data are grouped in different clusters free from preconceptions. Unlike other clustering methods, the partitioning around medoids (PAM) model is a unique and flexible clustering method that allows the entry of various forms of variables (nominal, ordinal, or scalar) besides numeric variables. ^{17 , 18} In the PAM model, one case is selected as a medoid, and the data with the least dissimilarities to the medoid are clustered around it. The medoids are then replaced to gain the smallest dissimilarity. This process continues until no change takes place in medoids and the determined number of clusters itself. This unique process leads to separated data groups with the least within-group dissimilarity and maximum between-group dissimilarity. ¹⁹ This highly accurate method classifies patients into clusters based on their true dissimilarities and distinguishes the important variables for clustering. Thus, classifications based on this model would be simple, most time-efficient, and require less clinical examination variables.

Studies evaluating the reliability of the MSI model have reported difficulties in classifying patients based on the signs and symptoms. ^{7 , 14} We believe that this could be due to difficulties in the judging part of the MSI model. Another study also reported that knee examination based on the MSI model is too time-consuming (about 45 minutes). ¹⁴ To address these issues, the main objective of the present study was to use the PAM clustering method for the classification of patients based on their signs and symptoms derived from the MSI model. In addition, we aimed to identify those key tests directly related to knee examination for the sole purpose of reducing examination time.

Materials and Methods

In the present cross-sectional study, patients with knee pain were recruited from several orthopedic and rehabilitation clinics affiliated to Shiraz University of Medical Sciences (Shiraz, Iran) during February-December 2018. The participants were selected using the convenient sampling method, and the sample size was estimated in accordance with previous studies, ^{5 , 14} and by using the statistical rules of thumb. ²⁰ A total of 209 patients with knee pain referred for physical therapy by an orthopedic surgeon were recruited in the study. The study was approved by the local Medical Ethics Committee (number: IR.SUMS.REC.1396.S993), and written informed consent was obtained from the participants.

The inclusion criteria were aged 18-60 years and experiencing non-traumatic pain around the knee (scoring between 3 and 7 in a standing position on a numerical rating scale) for the last two months. ^{5 , 14} The exclusion criteria were any history of surgery (bone osteotomy, bone fracture repair or surgical correction of structural deformities in the trunk or lower extremity), major general metabolic or systemic diseases, neurological diseases (radiculopathy), obvious leg length discrepancy leading to limping, pregnancy; and the use of walking aids, analgesic, and anti-inflammatory drugs up to or at the day of examination. ^{5 , 14} The examiner was a physical therapist with 12 years of experience in treating patients using the MSI model. All procedures were conducted according to the Declaration of Helsinki.

The examination procedure began with a documentation of the patient’s history (height, weight, age, sex, pain location, and intensity) followed by a physical examination. The activity level was measured using the Persian version of the Tegner questionnaire, which was previously validated with an acceptable level of reliability. ²¹ The questionnaire is a self-administered activity rating system (based on a scale of 0 to 10) for patients with various knee disorders. The physical postural examination included assessment of correct alignment (anterior, posterior, and side views) in standing and sitting positions as described in the MSI model. ¹⁰ The knee, femoral, and tibial movements were evaluated during different activities such as walking, half-squatting, single-leg stance (comparing both sides), and stair ambulation. A walkway and an adjustable chair were used to perform the walking and sit-to-stand transfer, respectively. Climbing was performed on stairs of 20 cm height.

The initial perceived level of pain was recorded during each movement. In case of faulty movement patterns, the therapist trained the participants on the correct pattern, requesting repetition of the movement, and report of the perceived pain level. The difference in the pain level between faulty and corrected movements was coded for each activity and rated from 0 to 7 (0: no change, 1: valgus correction, 2: varus correction, 3: hyperextension correction, 4: patellar movement correction, 5: no immediate correction available, 6: no alignment deficit, and 7: increased pain). This process was repeated in all positions. The examiner palpated each segment during movements, and if an activity was performed incorrectly (with valgus or excessive tibial external rotation), the participant was informed about the correct movement pattern. The movement was then repeated in the corrected form (if possible) and re-evaluated by the examiner. If an obvious difference was found between the previous and corrected movements, the faulty movement was recorded as the impairment. Impairments that could not be corrected immediately (hypomobility) were recorded without repetition of the movement. The deployed coding system was based on the MSI model and previous studies. ^{5 , 14} A total of 41 scalars (pain) and nominal (alignment deficits) variables were counted during different activities. A muscle stiffness-flexibility test was performed in the supine position to determine the muscle-joint relative flexibility. It consisted of a two-joint hip flexors length test, and the result was reported positive if the tibia was abducted or rotated laterally while lowering the hip in extension. The joint integrity was measured by assessing the accessory motions of the patellofemoral and tibiofemoral joints. The McConnell patellar test ¹⁰ and patellar grind test were also performed. The foot arch was measured by determining the Feiss line and classified as high, low, or normal arch. The variables collected during examination based on the PAM clustering analysis are presented in table 1. Manual muscle testing was also performed for the muscles of the lower extremity. However, the results were not used in the PAM analysis, since muscle power is not a discriminating factor for clustering.

The extracted data from all patients were analyzed using ClusterR and VarSelLCM packages of R software (R core team, version 3.5.3, New Zealand). PAM analysis allows the management of different types of variables (nominal, ordinal, or scalar) and is robust against outlier observations. ^{11 , 17} Moreover, it reduces the number of variables, prioritizes, and categorizes data into homogeneous clusters. The weight of each variable was defined through the discriminatory power (DP), i.e., the contribution of each variable in cluster analysis. ^{11 , 17} After the initial PAM clustering, the analysis was repeated for each cluster to identify sub-clusters that included all variables. All other statistical analyses were performed using SPSS software, version 25.0 (IBM Inc. Chicago, IL, USA). The quality of the clusters and sub-clusters was analyzed using the silhouette coefficient internal index, ranging from 1 to -1. Values close to 1 indicated more dense clusters with more distances from other clusters (i.e., best possible categorization) and those close to -1 indicated too many or too few clusters. ²²

Results

A total of 209 patients (57 men, 152 women) participated in the study of which 108 (51%) had pain of the right knee and 101 (48%) had pain of the left knee. The mean age of the patients was 43±12 years, the mean weight of 73±13 kg, and the mean height of 166±8 centimeters. Based on a visual analog scale, the pain level at the time of examination in standing position was 17±19. The mean total score of the Tegner questionnaire was 4±1. Based on all signs and symptoms, the PAM clustering analysis resulted in two main clusters and six sub-clusters.

Main Clusters

The optimum number of clusters was two with a silhouette coefficient value of 0.41. The results of the PAM clustering analysis showed that 23 out of the 41 variables were necessary for an initial clustering of patients into two main clusters (table 2). The clusters were labeled according to the most discriminative features of the diagnosed condition. The main discriminative characteristic was knee valgus, thus the corresponding clusters were labeled “Valgus” and “Non-valgus” The non-valgus cluster included characteristics of varus or no deformity. The valgus and non-valgus clusters contained 120 (57.4%) and 89 (42.6%) patients, respectively. Among all variables, the frontal plane knee deficits (valgus and varus) had the highest DP, especially in activities such as sit-to-stand transfer (DP=90.76), stair ambulation (DP=84.82), and partial squat (DP=77.10). The PAM analysis was repeated for each cluster to determine sub-clusters:

Sub-clusters of Valgus

The optimum number of sub-clusters of valgus was two with a silhouette coefficient of 0.26. The results of the PAM analysis showed that 12 out of the 41 variables were necessary for the sub-clustering of patients (table 3). Two sub-clusters were selected based on the presence or absence of knee hypomobility, being the most discriminative characteristic among all sub-clusters. The characteristics of each sub-cluster are listed in table 3. These sub-clusters were labeled “Valgus with hypomobility” and “Valgus”, since the most discriminative feature was knee hypomobility with the highest DP, especially in a single-leg stance (DP=65.30), walking (DP=57.71), and tibiofemoral joint integrity (DP=48.48).

Valgus with Hypomobility: A total of 16 (7%) patients with a mean age of 52 years exhibited valgus combined with hypomobility. Movement pattern modification, especially in the partial squat and stair ambulation, was the major correction to alleviate the symptoms. The patients also exhibited knee valgus in the sit-to-stand transfer. The major discriminator was the lack of knee extension, especially in a single-leg stance (DP=65.30) and walking (DP=57.71).

Valgus: A total of 104 (49.7%) patients with a mean age of 41 years exhibited valgus in a sit-to-stand transfer, partial squat, and stair ambulation as major discriminators.

Sub-clusters of Non-valgus

The PAM analysis was repeated for the non-valgus cluster to identify its sub-clusters. The optimum number of sub-clusters was four with a silhouette coefficient of 0.25. The results showed that 23 out of the 41 variables were needed for the sub-clustering of patients (table 4). These sub-clusters were labeled “Hyperextension”, “Hypomobility”, “Varus”, and “Minimal pain and dysfunction”. Among all the variables, joint integrity (DP=41.49), symptoms in single-leg stance (DP=38.24), symptoms in stair ambulation (DP=36.77), and knee hyperextension and hypomobility in single-leg stance (DP=34.04) had the highest DP.

Hyperextension: A total of 10 (4%) patients with a mean age of 32 years were clustered as having hyperextension, especially in single-leg stance and walking. Although correction of faulty movement pattern may not immediately reduce the symptoms in these patients, attention to the signs of sagittal-plane knee function impairments can be a discriminator for categorization since all patients had hyperextension in walking (DP=34.01) and standing on the examined leg (DP=34.04).

Hypomobility: A total of 23 (11%) patients with a mean age of 54 years exhibited a lack of full extension, especially in single-leg stance and walking. Movement pattern modification reduced the pain, particularly in stair ambulation and single-leg stance. All patients exhibited a lack of knee extension in stair ambulation (DP=28.96). Most of the patients with hypomobility exhibited varus in movements, especially in sit-to-stand transfer (73%), partial squatting (73%), and stair ambulation (78%).

Varus: A total of 26 (12.4%) patients with a mean age of 48 years exhibited knee varus as the main sign. The majority of these patients (73.1%) immediately showed reduced symptoms after movement pattern correction, especially in single-leg stance.

Minimal Pain and Dysfunction: The last sub-cluster consisted of 30 (14.3%) patients with a mean age of 35 years who had no preferred alignment fault and movement pattern correction or modified their preferred movement pattern to relieve pain symptoms. Most of them were young athletes (Tegner score of 5) scoring 8 out of 100 on a visual analog scale for average pain. The manual muscle testing of these patients showed a higher muscle strength and lower pain intensity than other patients.

A total of 23 examination variables were essential in sub-clustering knee classifications. 18 items were not considered group discriminators although they could have contributed to patient management. Irrelevant items included factors related to prone position (symptoms, tibial abduction-adduction, tibial medial-lateral rotation), tibial medial-lateral rotation (walking, partial squat, sit-to-stand transfer, single-leg stance, knee flexion while standing on the uninvolved limb, and stair ambulation); the McConnell patellar test, foot pronation, and femoral medial-lateral rotation in single-leg stance position.

Discussion

In the present study, we found that 23 out of the 41 clinical examination variables were adequate to categorize the patients in appropriate clusters and sub-clusters. We identified two main clusters (valgus and non-valgus) and six sub-clusters (valgus, valgus with hypomobility, hyperextension, hypomobility, varus, and minimal pain and dysfunction). The results showed that patients with knee impairment can be classified efficiently with a fewer number of examination variables than the MSI model. These variables were prioritized according to importance and DP in each step. The PAM analysis was also used as a novel method for knee examination to allocate patients to different sub-clusters.

A comparison of PAM sub-clusters with the MSI model showed that both models used similar impairments (valgus, varus, hypomobility, and hyperextension) as classification criteria. Patients with knee extension syndrome, as classified in the MSI model, were not entered into the study due to the reported low presence. ^{5 , 14} In line with the MSI model, patients in the sub-cluster “minimal pain and dysfunction” mostly exhibited the characteristics of patients with patellofemoral pain and knee joint hypermobility. These patients did not express pain during routine tests, making it necessary to perform more specific high loading or neuromuscular tests to determine the characteristics of this cluster. Patients with no specific alignment deficits have been labeled as neuromuscular or muscular deficits, one of the subcategories of patellofemoral pain. ²³ Patellofemoral dysfunctions were not classified as a separate cluster for two reasons. First, tests such as stair climbing were not effective in discriminating the pain source of the patellofemoral or tibiofemoral joint. ²⁴ Nonetheless, they were among the most important discriminative variables in our study. Second, pain variables were not that essential in diagnostic tests for the patellofemoral joint impairment. ²⁴ Keays and colleagues categorized patients with patellofemoral pain into four main categories. Among all, the patellar osteoarthritis group was only distinguished by the use of X-ray imaging. ²⁵ In the current study, patellar deficiency associated with knee impairments was assigned to the main sub-clusters.

The majority of our patients had dynamic knee valgus as the dominant movement pattern (valgus cluster). This was in line with previous studies reporting a high frequency of valgus movement in patients with knee pain. ^{14 , 25} The main clusters identified in the current study were somehow similar to previous studies. ^{5 , 14 , 25}

Signs can be a key factor in classifying patients when symptoms are not conclusive. We found that signs were more important in primary clustering and valgus sub-clustering of patients since, they were more obvious and with higher DP than symptoms. However, symptoms had a higher diagnostic value in patients with a dominant movement pattern other than valgus. Kajbafvala and colleagues also reported similar findings. ⁵ They used factor analysis to validate four syndromes of the knee classification system in which only one factor included symptoms and the other three included signs only. Salsich and colleagues found that despite the correction of tibiofemoral alignment, the patient’s pain did not decrease significantly. ¹³ Another study on the validation of the knee MSI model reported that the category “patellar lateral glide syndrome” was the only factor in which pain was important. ⁵ This syndrome was not identified in the PAM sub-clustering, since the examination variables related to the patella (pain location, McConnell patellar and symptom alleviation with patellar correction) did not have a DP higher than other variables. On the other hand, deficient patella rarely occurs in isolation and is usually a secondary diagnosis based on a primary diagnosis of tibiofemoral rotation or knee hyperextension. ^{11 , 13} Additional tests such as X-ray or trunk evaluation could be useful to attain a more uniform group. ^{25 , 26}

In the present study, we also identified a pattern for knee examination based on the most important variables. Physical therapists should pay attention to the presence of valgus in activities, since it is the most observed faulty movement pattern among patients. It is very common to diagnose this pattern in patients during sit-to-rise transfer, stair ambulation, or partial squat. After the diagnosis of knee valgus in conjunction with hypomobility, the patient may be treated differently from those without such impairment. Hypomobility can be easily detected when a patient stands on the impaired leg or walks. Back again to the first step, if the patient shows a varus pattern in movement or has no frontal plane deficits, therapists should pay attention to the presence of deformities in the sagittal plane such as hypomobility or hyperextension. In this step, therapists should primarily examine the joint, pay attention to the pain pattern, and observe the faulty movements especially in walking and standing on the impaired leg. If therapists need a complete and accurate assessment, all the 23 variables should be examined. The identification of the applicable patient’s cluster is determined by the skills of the physical therapist on examination, palpation, and interpretation of the findings.

The present study had some limitations. First, patients with severe symptoms (numerical rating scale above 7 in standing position or failure to perform the test because of severe pain) were not included due to the inability to perform all the tests, especially those with high loading activities. Second, clustering analysis could not classify those patients showing high tolerance to tests, since they could not fully reveal deficits in the signs and symptoms of a test. Usually, activities with high physical demand (single-leg hop test and single-leg squat test) are performed for accurate classification. ¹⁴ Another limitation of the study was that we applied tests that can generally be used for all patients with knee pain. Hence, specific tests such as single-leg squat, jump, and hop tests were not performed on patients with a low level of impairments.

Conclusion

A new direct knee examination method is introduced that organizes important discriminative tests and requires fewer clinical examination variables. By examining the essential variables determined in each step, one can easily classify patients with knee impairments using the proposed knee examination approach. Future studies should focus on comparing the result of the current study with the syndromes described in the MSI model. To the best of our knowledge, no studies have been performed on the efficiency of the MSI model on patients with knee pain. Since disabled patients with acute traumatic or disability-related pain were excluded, future studies should consider the inclusion of such patients.

References

1. Kraus T, Svehlik M, Singer G, Schalamon J, Zwick E, Linhart W. The epidemiology of knee injuries in children and adolescents. Arch Orthop Trauma Surg. 2012; 132:773-9. DOI | PubMed
2. Pereira D, Peleteiro B, Araujo J, Branco J, Santos RA, Ramos E. The effect of osteoarthritis definition on prevalence and incidence estimates: a systematic review. Osteoarthritis Cartilage. 2011; 19:1270-85. DOI | PubMed
3. Nguyen US, Zhang Y, Zhu Y, Niu J, Zhang B, Felson DT. Increasing prevalence of knee pain and symptomatic knee osteoarthritis: survey and cohort data. Ann Intern Med. 2011; 155:725-32. Publisher Full Text | DOI | PubMed
4. Harris-Hayes M, Sahrmann SA, Norton BJ, Salsich GB. Diagnosis and management of a patient with knee pain using the movement system impairment classification system. J Orthop Sports Phys Ther. 2008; 38:203-13. DOI | PubMed
5. Kajbafvala M, Ebrahimi-Takamjani I, Salavati M, Saeedi A, Ashnagar Z, Pourahmadi MR, et al. Validation of the movement system impairment-based classification in patients with knee pain. Man Ther. 2016; 25:19-26. DOI | PubMed
6. Fritz JM, Delitto A, Erhard RE. Comparison of classification-based physical therapy with therapy based on clinical practice guidelines for patients with acute low back pain: a randomized clinical trial. Spine (Phila Pa 1976). 2003; 28:1363-71. DOI | PubMed
7. Van Dillen LR, Sahrmann SA, Norton BJ, Caldwell CA, McDonnell MK, Bloom NJ. Movement system impairment-based categories for low back pain: stage 1 validation. J Orthop Sports Phys Ther. 2003; 33:126-42. DOI | PubMed
8. Karayannis NV, Jull GA, Hodges PW. Physiotherapy movement based classification approaches to low back pain: comparison of subgroups through review and developer/expert survey. BMC Musculoskelet Disord. 2012; 13:24. Publisher Full Text | DOI | PubMed
9. Whatman C, Hume P, Hing W. The reliability and validity of physiotherapist visual rating of dynamic pelvis and knee alignment in young athletes. Phys Ther Sport. 2013; 14:168-74. DOI | PubMed
10. Sahrmann S. Movement system impairment syndromes of the extremities, cervical and thoracic spines-e-book. Amsterdam: Elsevier Health Sciences; 2010.
11. Ludewig PM, Lawrence RL, Braman JP. What’s in a name? Using movement system diagnoses versus pathoanatomic diagnoses. J Orthop Sports Phys Ther. 2013; 43:280-3. DOI | PubMed
12. Willis S, Rosedale R, Rastogi R, Robbins SM. Inter-rater reliability of the McKenzie System of Mechanical Diagnosis and Therapy in the examination of the knee. J Man Manip Ther. 2017; 25:83-90. Publisher Full Text | DOI | PubMed
13. Salsich GB, Graci V, Maxam DE. The effects of movement pattern modification on lower extremity kinematics and pain in women with patellofemoral pain. J Orthop Sports Phys Ther. 2012; 42:1017-24. Publisher Full Text | DOI | PubMed
14. Kajbafvala M, Ebrahimi-Takamjani I, Salavati M, Saeedi A, Pourahmadi MR, Ashnagar Z, et al. Intratester and intertester reliability of the movement system impairment-based classification for patients with knee pain. Man Ther. 2016; 26:117-24. DOI | PubMed
15. Bierma-Zeinstra SM, Bohnen AM, Bernsen RM, Ridderikhoff J, Verhaar JA, Prins A. Hip problems in older adults: classification by cluster analysis. J Clin Epidemiol. 2001; 54:1139-45. DOI | PubMed
16. Selfe J, Janssen J, Callaghan M, Witvrouw E, Sutton C, Richards J, et al. Are there three main subgroups within the patellofemoral pain population? A detailed characterisation study of 127 patients to help develop targeted intervention (TIPPs). Br J Sports Med. 2016; 50:873-80. Publisher Full Text | DOI | PubMed
17. Kassambara A. Practical guide to cluster analysis in R: Unsupervised machine learning. London: Sthda; 2017.
18. Lesmeister C. Mastering machine learning with R. Birmingham: Packt Publishing Ltd; 2015.
19. Verma R, Puntambekar D. Comparison of partitioning algorithms for categorical data in cluster. Int J Eng Sci. 2018; 18701
20. Van Belle G. Statistical rules of thumb. New Jersey: John Wiley & Sons; 2011.
21. Negahban H, Mostafaee N, Sohani SM, Mazaheri M, Goharpey S, Salavati M, et al. Reliability and validity of the Tegner and Marx activity rating scales in Iranian patients with anterior cruciate ligament injury. Disabil Rehabil. 2011; 33:2305-10. DOI | PubMed
22. Rousseeuw PJ. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of computational and applied mathematics. 1987; 20:53-65. DOI
23. Witvrouw E, Werner S, Mikkelsen C, Van Tiggelen D, Vanden Berghe L, Cerulli G. Clinical classification of patellofemoral pain syndrome: guidelines for non-operative treatment. Knee Surg Sports Traumatol Arthrosc. 2005; 13:122-30. DOI | PubMed
24. Stefanik JJ, Neogi T, Niu J, Roemer FW, Segal NA, Lewis CE, et al. The diagnostic performance of anterior knee pain and activity-related pain in identifying knees with structural damage in the patellofemoral joint: the Multicenter Osteoarthritis Study. J Rheumatol. 2014; 41:1695-702. Publisher Full Text | DOI | PubMed
25. Keays SL, Mason M, Newcombe PA. Individualized physiotherapy in the treatment of patellofemoral pain. Physiother Res Int. 2015; 20:22-36. DOI | PubMed
26. Nakagawa TH, Maciel CD, Serrao FV. Trunk biomechanics and its association with hip and knee kinematics in patients with and without patellofemoral pain. Man Ther. 2015; 20:189-93. DOI | PubMed