PFP sub-groups based on pelvic acceleration
The first purpose of the present study was to determine if running gait patterns in individuals experiencing PFP at the time of testing could be clustered into homogeneous sub-groups based on combinations of pelvic acceleration components. In support of our hypothesis, two distinct and homogenous sub-groups (clusters) were present in females with PFP, and these clusters were different when compared to PFP males. These results are similar to previous studies that also reported two to three different running patterns based on visual inspection of 3D kinematic data [
6,
7] and mechanical differences between males and females with PFP [
8].
There were no significant differences in running speed between sub-groups, which is a factor that has been shown to affect axial segment acceleration [
34], especially in the ML axis [
35]. Male subjects were significantly taller and heavier than females and these anthropometric differences are known to influence 3D kinematics during running [
36]. However, there was a very weak correlation for height, and no correlation for body mass with the acceleration PCs that presented differences between sub-groups suggesting that the relationship with those factors was minimal.
The advantage of investigating pelvic acceleration as a measure of running mechanics is that it is less influenced by marker placement errors and is a much simpler method than a full 3D gait assessment, as it depends only on the trajectory of a single pelvic marker cluster. Additionally, these factors allow for the use of data from multiple research centres, allowing for the application of ‘big data’ analytics and a better understanding of the interaction between biomechanical factors and musculoskeletal injuries [
10,
11]. Furthermore, the results of the present study opens the possibility for the use of wearable devices for data acquisition, such as a single triaxial accelerometer on the pelvis, an approach which is becoming increasingly popular in industry and health care [
18,
20]. Therefore, the current work identifying sub-groups of PFP patients is a novel finding that can guide future studies in providing better context that can hopefully improve clinical practice.
Identification of differences in running gait patterns between sub-groups
A secondary purpose was to analyze peak joint angles between clusters to better understand the practical and clinical implications of clustering subjects with PFP based on 3D pelvic acceleration data. In general, differences in joint kinematics were sex-related, since there were no significant differences between female clusters, except for peak hip internal rotation. Moreover, the magnitude of mean differences were within the threshold for detectable kinematic changes reported by Osis et al. [
15] for knee abduction (3.4
o) and hip internal rotation (5.6
o). However, the differences in ankle eversion and hip adduction between males and females are greater than the error margins caused by marker placement errors, confirming the findings of Willy et al. [
8] who reported males with PFP to have less hip adduction than their female counterparts.
Phinyomark et al. [
12] reported the existence of two different sub-groups of asymptomatic runners based on a HCA of lower limb joint kinematics, and when they compared the peak knee abduction angles of those clusters with a sample of subjects with PFP, group differences were dependent on the cluster of healthy individuals that was used as reference. Interestingly, all PFP sub-groups from the current study presented greater values of knee abduction when compared to the ones reported for healthy runners (healthy C1: 8.0
o; healthy C2: 4.4
o). However, there is a tendency for a progressively greater alteration in knee frontal plane angles when comparing males to females in C1 and C2, although there was no significant difference between the female clusters. This could be related with distinct pathomechanical pathways or differences in response to treatment. For example, in a previous work [
37] we found that non-responders to exercise treatment protocol presented greater knee abduction angles during late stance and swing phases of running gait, and the current findings suggest that this could be identified by pelvic acceleration data.
To our knowledge, this is the first study to investigate pelvic acceleration profiles in runners with PFP, and the identification of sub-groups could generate insights about differences in pathomechanics or adaptations to pain. Additionally, the analysis of segmental acceleration profiles minimizes measurement imprecisions originating from marker placement errors that propagate into the calculation of joint angles in 3D kinematics [
14,
15]. Furthermore, the results of the current study suggest that accelerations acquired using wearable devices [
24] may utilise this method in a clinical setting as an evidence-informed method to improve patient care and rehabilitation decisions.
The pelvic acceleration data can provide some clinical insight that can help clinicians make decisions regarding treatment options. For example, peak resultant pelvic acceleration is related to center of mass acceleration during 10 to 75% of stance phase [
38]. Therefore, pelvic accelerations can provide some insights on shock absorption and lower limb stiffness. Nevertheless, this connection must be made with caution, since accelerations based on segmental measures overestimate the behavior of center of mass [
38]. Women in C2 presented a higher VT peak acceleration, suggesting a diminished capacity for shock absorption. Since no differences in peak knee flexion angles were detected, this could be an indication of greater leg stiffness in these subjects, which is partially supported by the findings that women present higher leg stiffness during running [
39] and drop jump landing tasks [
40] when compared to males. In contrast, females in C1 were similar to males regarding VT acceleration patterns, which could be explained by the lower VT displacement.
Women also presented higher and delayed peak accelerations in the ML direction, suggesting differences in the control of side-to-side body movement during the first half of the stance phase, when these oscillations occur. This pattern could be related to the larger hip adduction angles exhibited during running, which led to increases in ML accelerations. In addition, females in C2 displayed a delay in peak AP accelerations in early stance, causing a prolonged period of deceleration. It is possible that this finding is related to strength differences between males and females [
41,
42], as stronger individuals may be able to exert shorter impulses to achieve the same net change in momentum, however, strength differences were not quantified in the current study.
Although the identification of sub-groups among the female subjects with PFP did not coincide with significant differences in peak lower limb joint angles, there seems to be a progression of values in knee abduction and hip internal rotation depending on the cluster of female subjects. Specifically, there is a tendency for C1 to have lower knee abduction and higher hip internal rotation than C2. These factors could be related to symptom severity or differences in response to treatment, but would need further investigation.
Limitations
In addition to the differences in height and weight between males and females that were already discussed, other limitations to the current research study are acknowledged. First, this study included both subjects with uni- or bilateral involvement and with secondary pain symptoms besides PFP, which could have also modified running mechanics. However, there was no significant difference in the distribution of those variables between the two subgroups, leading us to believe that it was not an important factor for this clustering. Additionally, these types of patients are frequently seen in clinical practice, therefore these PFP patients are important to include in research studies.
Second, we did not have access to other clinical variables that could influence running mechanics and explain the differences that were found between sub-groups. For example, Selfe et al. [
43] has described 3 clusters of PFP patients that were grouped based on clinical measures of strength, flexibility and joint alignment and mobility. Additionally, experimental pain induction in the knee joint has been shown to cause reductions in peak torque in maximal voluntary contraction of knee flexors and extensors [
44] and increased sway displacement during quiet stance [
45], indicating that pain level could be a driver of changes in motor control. Therefore, future studies should include the aforementioned clinical variables to investigate whether they are related to the differences in running pattern found between sub-groups to have a better understanding in a clinical context.
Finally, this investigation used an HCA approach, which is an unsupervised machine learning technique suitable for exploratory analyses, to determine whether this type of data could be useful in the identification of subgroups within a cohort of runners with PFP. Overall, our hypothesis was supported by the findings and suggest that a supervised analysis could also be applied to identify specific subgroups with specific clinical relevance. For example, recent work from our laboratory used a supervised machine learning method to classify runners with PFP into responders or non-responders to exercise treatment based on running kinematic data, achieving 78% of classification accuracy [
37]. Thus, a similar approach could be applied in this context, using pelvic acceleration data to develop an objective method for the identification of such subgroups with greater accessibility in a clinical setting. Regardless, the present study is an important first step to verify the utility of simple measures, like pelvic accelerations, for the objective assessment of gait biomechanics.