Systematic Review on the Applicability of Principal Component Analysis for the Study of Movement in the Older Adult Population

Principal component analysis (PCA) is a dimensionality reduction method that has identified significant differences in older adults’ motion analysis previously not detected by the discrete exploration of biomechanical variables. This systematic review aims to synthesize the current evidence regarding PCA use in the study of movement in older adults (kinematics and kinetics), summarizing the tasks and biomechanical variables studied. From the search results, 1685 studies were retrieved, and 19 studies were included for review. Most of the included studies evaluated gait or quiet standing. The main variables considered included spatiotemporal parameters, range of motion, and ground reaction forces. A limited number of studies analyzed other tasks. Further research should focus on the PCA application in tasks other than gait to understand older adults’ movement characteristics that have not been identified by discrete analysis.


Introduction
The older adult population, those with age above 60 years, will increase in the coming decades in Europe [1]. Accounting for almost half of the total health costs [2], the aging population will increase the burden on healthcare services [3] expressed through an estimated increase from 3% to 4% in gross domestic product from 2004 to 2050 [4].
Biological aging can be defined as the progressive loss of function and represents a constant decrease in multisystemic capacity [5][6][7] that can be expressed in changes in movement patterns in various tasks [8]. Different biomechanical analyses have been associated with fall risk in older people, such as the displacement of the center of pressure in the standing position [9], decreased speed, stride length, and single limb support time in gait [10]. Other movement modifications were identified due to the aging process, such as decreased lumbar range of motion [11] during sit-to-stand, an indicator of functional independence in daily life [12]. However, because these findings have been obtained from discrete approaches, such as descriptive statistics and statistical inference based on only some parameters of the waveform [10,11], there is a consistent risk of information loss [13]. Additionally, other clinical measures assessed by classical psychometric procedures may have led to dubious conclusions [14], while others discarded several parts of the information and required a greater number of trials from participants before drawing conclusions [15]. Advanced multivariate analysis and machine learning methods have been applied to fully translate the complexity of the interactions between the variables [13]. Principal component analysis (PCA) is a multivariate statistical technique that reduces the volume of data to a smaller number, considering all the information

Materials and Methods
This systematic review was carried out according to the Preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement 2020 [30], and was registered on the PROSPERO (International Prospective Register of Ongoing Systematic Reviews) platform with the registration number CRD42022329200 (https://www.crd.york.ac.uk/ prospero/display_record.php?RecordID=329200, accessed on 23 October 2022).

Eligibility Criteria
According to the PICOS strategy, the criteria defined for inclusion of the studies were: population, older adults aged 60 years or over, living in the community; exposure, studies that applied the PCA in the biomechanical analysis of the movements of older adults; movement analysis associated with the aging process; comparator, young adults (less than 60 years); outcomes, tasks and biomechanical variables (kinematics and kinetics) and, if described, the instruments used to collect the biomechanical data; study design, observational studies (cross-sectional and longitudinal). As exclusion criteria, it was decided to exclude the studies if their participants were institutionalized (or if the data from institutionalized participants could not be separated from community-dwelling ones) or if the participants presented some major pathology with repercussions on the performance of movement, such as stroke or Parkinson disease (or if the data from participants with these conditions could not be separated from the complete study sample). The search was restricted to the studies published in the Portuguese and English languages, that were available in the last 20 years.

Selection and Data Collection Process
The studies were searched using three databases: MEDLINE (PubMed), Scopus and Web of Science (Supplementary Materials). Specific search algorithms were elaborated for each database, as described in the Table 1 for Pubmed and for the other two databases in the Supplementary Materials. Each concept was searched according to the database search instructions, using the MeSH terms and synonyms. Two reviewers independently assessed the studies' titles and abstracts in the identification phase. Then, in the screening phase, the same reviewers assessed the full texts. Disagreements about whether a study should be included were resolved if there was an oversight of information on the part of one of the reviewers or by discussion, consulting a third reviewer in the cases of different interpretation of studies content. The two reviewers used a pre-defined table to extract data from the included studies.

Assessment of Methodologic Quality
The study design of the included studies is observational, and as there is no gold standard of risk of bias (ROB) tools for observational studies, different tools have been used in previous systematic reviews [31]. The most commonly used ROB tools are the Newcastle-Ottawa Scale (NOS) and the Downs & Black scale [31]. Downs & Black scale was found to be fairly comprehensive; easy to use and clear descriptions of how to score items [32]. The scale has good test-retest reliability (r = 0.88), good interobserver reliability (r = 0.75) and a high internal consistency (KR-20:0.89) [33]. Accordingly, it was decided to evaluate the ROB of articles included in this review by two reviewers. As in the previous processes, the differences between the two reviewers were solved by consulting a third reviewer. The Downs & Black instrument consists of 27 items that assess the quality of the study, including data reporting, external validity, internal validity (bias), internal validity (confounders) and power [33]. In this study, a modified version of the Downs & Black scale, adapted by Rollo et al., 2020, was used, in which the authors removed 10 items (8, 13-15, 17, 19, and 21-24) from the original scale, because they were considered not relevant for the analysis of observational studies. In addition, the authors modified the items (4, 5, 9, 26 and 27) and created two new items, one of which describes the criteria of internal validity and the other is related to the power of study [34]. The modified Downs & Black scale is then composed of 19 items, and the possible score on each item is 0 or 1. The maximum possible score is 19 points (all positive signs), with a higher score indicating higher quality [34].

Characterization of the Included Studies
The identification, main purpose and conclusions of the 19 included studies is presented in Table 2 The publication year of each study varied between 2002 [50] and 2021 [37,49], and out of the nineteen studies, eighteen were cross-sectional, and one classified

Characterization of the Included Studies
The identification, main purpose and conclusions of the 19 included studies is presented in Table 2. The publication year of each study varied between 2002 [50] and 2021 [37,49], and out of the nineteen studies, eighteen were cross-sectional, and one classified themselves as a retrospective cohort [39]. The study's sample size ranged from 14 [46] to 239 [54] participants, constituting a total of 1281 participants. The older group sample size ranged from seven [46] to 127 [54] participants. The mean age of the older adults group ranged from 63, 7 years old [49] to 79, 43 years old [46], while the younger groups' age ranged between 4.8 years [38] to 55,1 years old [54] (Table 3). Six studies did not present any information regarding the gender of the participants [36,40,41,44,47,53]. The mean percentage of female participants was 56.9% across the studies that reported this information. Moreover, there was one study that only included participants of the male gender [50], and one study that only included female participants [51]. Assess if the whole-body movement and/or motor control strategy differ as a function of age or sex in a forward reactive step to maintain balance.
PCA enabled to differentiate younger and older adults according to gender in terms of whole-body reactive stepping strategy and how ground reaction forces and kinetics support maintaining balance synergistically with whole-body movement strategy, when combined with multiple regression analysis.
The covariation remains stable between 15 and 70 years old.
Boyer & Andriacchi, 2016 [39] United States of America (USA) Assess the impact of age on knee function during walking in individuals with healthy knees as it applies to the development of knee osteoarthritis.
PCA analysis provided insight to the progressive changes in the magnitude of joint angles and in the kinematic coupling at the knee with age.

De Freitas et al., 2010 [40] Brazil
Assess age-related effects on postural responses following forward support surface translation throughout middle-adulthood and early old age.
Independent of age, the individuals were able to minimize center of mass backward displacements in response to forward perturbation and to revert the direction of this displacement at proper time with similar kinematics patterns. However, after the fifth decade changes in neuromuscular responses are observed.

Dewolf et al., 2019 [41] Belgium
Assess the effects of age on the intersegmental coordination in healthy young and elderly adults walking at matched speeds.
Older adults present decreased intersegment covariation with speed compared to young adults, mainly related to foot-shank coordination.

Gulde et al., 2019 [42] Germany
Assess the effects of speed of execution on upper-limb kinematics, in activities of daily living, with respect to age.
PCA revealed a movement strategy and age-dependent decline in primarily executive functions.

Kobayashi et al., 2016 [43] Japan
Assess age independent and most dominant sex differences observed in gait during normal walking.
PCA was able to identify a variation with significant age-sex interaction and another with significant sex difference but no age-effect or age-sex interaction.

Liu et al., 2020 [44] Taiwan
Assess the coordination of the multiple joints of the human body to maintain a stable posture and how it varies with age.
Aging increases the coupling strength and decreases the changing speed and the complexity of inter-joint coordination patterns. To understand equilibrium function and movement coordination in elderly by means of a whole-body goal-oriented task.
During whole-body movements, center of mass displacements are smaller in elderly compared to young adults and this postural aging effect is associated with straighter wrist paths. Despite these changes, high covariations of joint and elevation angles, observed in young adults, were also preserved in older adults.

Park et al., 2011 [46] USA
Assess age-related changes in finger coordination during accurate force and moment of force production tasks The magnitudes of the loading coefficients in the PC analysis suggested that the young subjects used mechanical advantage to produce moment while elderly subjects did not.

Reid et al., 2010 [47] Canada
To use PCA to compare the gait patterns between young and older adults during stair climbing The PCA and discriminant function analysis identified gait pattern differences between young and older adults.

Rosenblum et al., 2020 [48] Israel
To calculate total recovery time after different types of perturbations during walking and use it to compare young and older adults following different types of perturbations.
PCA showed differences in step length and step width recovery times between AP and ML perturbations.

Rowe et al., 2021 [49] Canada
To examine and describe age and sex-specific temporal pattern differences in lower extremity gait mechanics in asymptomatic adults.
The use of PCA enabled the observation of major sex-specific differences leading to the identification of an overall difference in walking gait strategy between healthy adult male and female participants, independent of age.

Sadeghi et al., 2002 [50] Canada
To identify the main structural characteristics of the sagittal knee muscle moment curves developed in elderly and young able-bodied subjects No significant differences were found between groups about the quality or magnitude of the sagittal knee peak muscle moment during the stance phase and early swing phase Slaboda, J. C., 2011 [51] USA To explore the influence of continuous visual flow, during and following a postural disturbance (i.e., support surface tilt), on the ability to reorient to vertical.
The fPCA revealed greatest mathematical differences in center of mass and center of pressure responses between groups or conditions during the period that the platform transitioned from the sustained tilt to a return to neutral position

Verrel et al., 2009 [52] Germany
To investigate the effects of concurrent cognitive task difficulty (n-back) on the regularity of whole-body movements during treadmill walking in women and men from 3 age groups.
Age seems to not influence gait regularity.

Wu et al., 2007 [53] China
To evaluate the use of Kernel-based Principal Component Analysis to extract more gait features (i.e., to obtain more significant amounts of information about human movement) and improve the classification of gait patterns.
Nonlinear gait features can be extracted to automatic classification of healthy young or older adults gait patterns.

Zhou et al., 2020 [54] Netherlands
To evaluate if different groups (healthy young-middle aged adults, healthy older adults, and geriatric patients) can be classified based on dynamic outcomes.

Tasks Assessed in Included Studies
Eleven studies assessed the gait task. Five studies assessed the task in treadmill under perturbed conditions [36,48], different speeds [38,41], and dual tasking [52], while the other six studies assessed the overground gait [39,43,49,50,53,54]. Three studies assessed upright standing in unperturbed [44] and perturbed conditions [40,51]. The remaining five articles assessed different tasks, such as stepping [37], preparing a cup of tea and a letter [42], grasping an object placed at the ground from a standing position [45], climbing stairs [47], and during maximal voluntary contraction of fingers [46] (Figure 2). five articles assessed different tasks, such as stepping [3 letter [42], grasping an object placed at the ground from a stairs [47], and during maximal voluntary contraction of

Methodological Quality Assessment
The methodological quality assessment score, according to the Downs & Black scale, ranged from ten [36,40] to 16 [49] and is presented in Table 4. The average score of the articles included is approximately 12.79 points, the fair level. Seven studies [37,39,43,[47][48][49]52] (36.8%) obtained a good classification and the remaining twelve (63.2%) obtained a fair classification. In general, the articles revealed very similar "Reporting" and "Internal validity" values, but only one scored in the "Power" section, while none scored in the "External validity" section.

Discussion
This systematic review aimed to summarize the tasks and biomechanical variables studied when PCA was applied in the study of the older adult population's movement compared with younger adults. The results of the systematic search reinforce the need to gather this information as this method has been widely used, mainly in Europe (n = 7) and North America (n = 8), and more than half of the included studies (n = 11) were published in the last decade.
The study's sample size ranged from 14 [46] to 239 [54] participants. In the literature, there is no agreement over the recommended sample size for the use of PCA, and the ratio between sample size and variables assessed. Guadagnoli and Velicer indicated that absolute minimum sample sizes, rather sample sizes as a function of the number of variables, are more relevant [55]. More recently, Osborne and Costello stated that both should be taken into consideration to avoid errors of interference, indicating that the best outcomes occur in analyses where large numbers of sample size and high ratios are present [56]. Because there is no simple method for calculating sample size in PCA [57], Comfrey and Lee (1992) cited by Osborne and Costello [56] suggested that "the adequacy of sample size might be evaluated very roughly on the following scale: 50-very poor; 100-poor; 200-fair; 300-good; 500-very good; 1000 or more-excellent". In the present review, 11 of the 19 articles included have fewer than 50 participants, so interpretation of their results should take that into consideration.
The most assessed task was the gait, with five studies assessing the task on a treadmill [36,38,41,48,52], while the other six studies assessed the overground gait [39,43,49,50,53,54]. Older adults' mobility can be influenced by multiple physiological and psychological factors [58], and other tasks as the balance should be assessed considering that its integrity is essential for activities of daily living efficacy [59]. Accordingly, only three studies assessed upright standing [40,44,51], all expressing promising results. Other activities, such as stair descent, which is regarded as one of the most difficult activities for older adults [60], are important to be assessed and processed in a broader context. In this review, only one study aimed to compare the gait patterns between young and older adults during stair climbing [47]. Therefore, there is a need to explore other tasks with the PCA approach. There are other activities, such as complex upper extremity-based manual activities of daily living tasks, in this review only assessed by Gulde et al. [42] which are still pending movement analyses based on kinematic markers [61], and consequently exploration by multivariate analysis.
PCA reduces the volume of data to a smaller number, and the visualization and statistical analysis of the new variables created, the principal components, can help to find similarities and differences between samples [16,63]. PCA was applied to extract features from several waveforms data [36][37][38][39][40][41][42][43][44][45][46][47][48][49][50]52]. In particular, PCA was used to analyze the angular covariance of the lower limb joints [38,40,41], extract space-time and kinematic data from gait [39,49], reduce the size of the data [37,47], and to assess motor coordination [36,42,44,45,52]. Additionally, the potential for PCA to uncover differences between groups was highlighted in three studies [43,48,50]. Accordingly, different applications of PCA were used within the included studies. Other studies used PCA variations, including functional PCA [51], and Kernel-based PCA [53,54]. Several included studies reported findings that were not possible by discrete analysis. Therefore, there is a need for the application of PCA in other tasks to understand older adult movement characteristics that have not been identified by discrete analysis. Cumbes and Azema proposed using the PCA to find feature patterns related to the autonomy-disability level, assessed by a disability scale, of elderly persons living in nursing homes [64]. In a longitudinal study, Shin et al. aimed to group diseases classified by the International Classification of Diseases using the PCA to extract comorbidity patterns and found that the principal component 1, which included diabetes, heart disease, and hypertension, was associated with an increased hazard ratio of mortality [65]. Some authors have already studied the kinematics of gait to cluster older adults with and without specific conditions [28,66,67]. The PCA clustering could be applied to kinematic and kinetic data of different daily performance tasks of community-dwelling older adults to cluster the autonomy-disability level and mortality. Early identification of those with disabilities and/or specific conditions could allow the introduction of prevention programs promoting older adults' independence.
The results of this systematic review should be analyzed considering that three databases were searched. However, the three databases chosen include a broader range of indexed studies. Another limitation of this review may be the lack of inclusion of studies in languages other than English and Portuguese. A wider language criterion could increase the number of included studies as multivariate analysis has been used worldwide.
Taking into account that the vast majority of studies applied PCA to the analysis of tasks such as gait, as stated previously, it is necessary to develop studies that would investigate other tasks, including other daily life activities. In the upcoming studies, it is also necessary to include larger sample sizes in order to fully take advantage of the potential of multivariate analysis. Furthermore, other structured reviews and meta-analyses aiming to understand the role of PCA in the biomechanical analysis of older adults, differentiating between individuals with diseases or conditions and healthy ones [28,66,67], would be beneficial, as the evidence in these topics grows.

Conclusions
The aim of this systematic review was to gather the current information related to the use of the PCA method in the study of movement in the older adult population. Accordingly, PCA has been applied globally, mainly in the study of gait and orthostatic position. The main variables assessed were spatiotemporal parameters, the range of motion of lower limb joints, and ground reaction forces. PCA was mostly used to analyze the angular covariance of the lower limb joints, extract space-time and kinematic data from gait, reduce the size of the data, and assess motor coordination. A limited number of studies analyzed other tasks. Therefore, considering the potential of multivariate analysis, further research should focus on the PCA application in tasks other than gait to understand older adults' movement characteristics that have not been identified by discrete analysis.
Supplementary Materials: The following supporting information can be downloaded at: https://www. mdpi.com/article/10.3390/s23010205/s1. Figure S1: Specific search algorithm for Scopus; Figure S2: Specific search algorithm for Web of Science; Figure