Postural Control

All of the beforementioned procedures in this section were carried out using PManalyzer software [38 (link)], a Matlab GUI specifically designed for PM variable computation (Matlab (R2019B), Natick, Massachusetts: The MathWorks Inc.). Only data from normal-walking trials was included in the independent variable computation. In short and in line with other recent endeavours using this approach [32 (link)–34 (link)], the data from all subjects and walking-speed trials were pooled into one matrix to allow for direct comparisons between subjects and across trials. Raw marker coordinates (XYZ) were first transposed so that each frame represents a posture vector, then centred by subtracting each posture vector from the mean posture vector to get postural deviations and normalized to their mean Euclidean distances to ensure an equal contribution by each subject/trial to the pooled dataset and subject-specific overall variance respectively. The data was also centred towards the centre-of-mass to create a body-position dependent coordinate system thus removing the potential inclusion of irrelevant body displacements from the PMs [23 (link)]. The pooled dataset of concatenated normal-walking trials accumulated into a 195,000 x 60 input matrix (100 Hz [Sampling rate] x approximately 50 seconds [Trial duration] x 3 [Three trials] x 13 [Number of subjects] x 60 [Marker coordinates]). The PCA algorithm decomposed this input into a covariance matrix of PM_k.
The PM_k within the output of this process were leave-one-out cross validated. The first 7 PMs presented a change of less than 15° when a participant was left out and were therefore included in further analysis [31 (link),38 (link)]. To compute the PA_k time-series, the PP_k time-series were first extracted as the PCA-scores derived from the PM eigen-vectors that represent deviations in posture across the orthonormal posture-space (25). The time-series were then further low-pass filtered using a 3^rd-order Butterworth filter at a cut-off of 10Hz and their spectral properties inspected with a Fourier analysis for frequency power above what is expected in noise free movement data [39 (link)]. No significant power was found at frequencies above 5Hz and so the effect of noise was deemed to have been sufficiently reduced. The control of PM_k was represented as the root-mean-square (RMS) of the PA_k time-series with respect to trial duration (t) (PA_kRMS) (Eq 1). In previous studies using this measure, static balance tasks were investigated where the accelerations in PA_k components could be directly related to postural control mechanisms [33 (link)]. During dynamic movements however, it must be noted that these accelerations additionally include dependencies such as walking-speed and so cannot solely reflect the actions of the neuromuscular control system [32 (link)]. Therefore, the PA_kRMS were then normalized by the respective walking-speed of the trial.

Postural control was evaluated using the path of COP. COP was measured on a force plate (9281E, Kistler, Winterthur) under three conditions: double-leg standing with eyes open (DEO); double-leg standing with eyes closed (DEC); and single-leg standing with eyes open (SEO). The order of each condition tested was chosen randomly. The dominant leg was selected for the SEO task. Leg dominance was determined after performing three trials of three functional tests [23 (link)]. First, subjects were asked to step onto a 40-cm box; the leg used to perform the step-up was identified as the dominant leg. Next, subjects were pushed from behind, and the leg that stepped out was identified as dominant. Then, subjects were asked to kick a soccer ball, and the leg used to kick the ball was recorded as the dominant leg. The leg that was dominant in two out of the three tests was considered the dominant leg for this study.
To attain the double- and single-leg stance position, subjects were instructed to keep their standing leg still, with their arms by their sides. For single-leg standing, they maintained the non-weight-bearing leg in a position of 90 degrees of knee flexion, keeping their thighs vertical. Before the test measurements were conducted, subjects practiced 3–5 trials of each test position. For the test measurements in each position, subjects were asked to stand for as long as possible, up to 30 seconds. The test was stopped when subjects were unable to maintain the requirements of the test position. The standing test measurements were performed once in each test position with a 10-second rest between each position, which were performed in a random order.
To process the data, the output from the force plate was introduced to the computer through an analog-to-digital converter (PH-770, DKH, Tokyo) at a sampling frequency of 1 kHz., and the COP trajectory was analyzed for 20 seconds, excluding the first 5 and last 5 seconds during double- and single-leg standing tasks. Fig 2 shows typical examples of the COP trajectory before and after the exercises. The following variables were used to describe the movement of the COP that were analyzed by TRIAS software (DKH Corp., Tokyo): total length (TL); mean velocity (MV = TL / total time); sway area (SA); maximum range of anteroposterior sway (AP range); maximum range of mediolateral sway (ML range); mean velocity of anteroposterior sway (AP velocity = AP length / total time); and mean velocity of mediolateral sway (ML velocity = ML length / total time). TL and SA were calculated with the following equation [24 (link)]:
$\mathrm{TL}={\displaystyle \sum _{n=1}^{N-1}\sqrt{{(A{P}_{n+1}-A{P}_n)}^2+{(M{L}_{n+1}-M{L}_n)}^2}}$
$\mathrm{SA}=\frac{1}{2}{\displaystyle \sum _{n=1}^N\left|A{P}_{n+1}M{L}_n-A{P}_{n}M{L}_{n+1}\right|}$
where N is the number of data points included in the analysis (20000 points) and n is the COP time series. The COP (AP and ML) time series were passed through a Butterworth low-pass digital filter with a 6 Hz cut-off frequency. TL is the total length of the COP trajectory; i.e., the sum of the distance between consecutive points of the COP trajectory. SA estimated by the area of a convex hull; the sum of the triangulation formed by two points on the COP trajectory necessary for calculating the convex hull. AP and ML ranges were the distance between the anterior and posterior peak displacements and the distance between the medial and lateral peak displacements.

We will examine the relationships between training predictors, neurobehavioral moderators and locomotor skill. The primary dependent variable is locomotor function (GMFM C for crawling and GMFM E for walking). The independent variables are error, movement index, postural control, and movement variability (see Table 1) which are measured repeatedly during training sessions Months 5–18 and will be subjected to data reduction. The moderators are early spontaneous movement behavior (Month 4), cognition (Month 5 for prone and Month 9 for upright), and motivation to move (Month 5 for prone and Month 9 for upright). One model will be developed for prone locomotor skill, and a second for upright locomotor skill. The small sample size and the potential correlation among the independent variables, including the moderators, will require careful examination and interpretation of the regression models' results. The focuses will be on obtaining estimates of slopes and their 95% CIs. We will explore modeling the dependent variable and the repeated measure of the independent variables by utilizing the mixed effects modeling approach. The three moderators may be grouped mean centered (i.e., subtracting group mean value from individual measure). The moderators will be examined individually in the mixed effect models. The moderating effects will be assessed by incorporating as an interaction of moderator X each of the independent variables utilized in the mixed effect model. The magnitude and the sign of the interaction term will be assessed to explore the moderating effects of the proposed moderators. Also, when appropriate, we will employ multivariate cluster analysis algorithms and the linear discriminant analysis to explore the relationship among and between variables. The results will help us to choose the important factors for estimating predicative models linking learning strategies to locomotor skill.

Locomotor function assessments will be completed for infants who progress to Phase 2 (the intervention phase) at six time points (Months 5, 7, 9, 12, 15 and 18). The primary outcome measure for Aim 1 is the Movement Observation Coding System (MOCS), a task-specific measure of locomotor performance. The MOCS uses video coding to assess postural control, arm and leg movements, and goal directed movement. The scale has 27 items with 5 items to assess position of head, trunk, hands and legs in prone and 22 items that assess the effect of the child's arm and leg movements on movement of the SIPPC and prone locomotion. The scale has been validated with infants with various disabilities (49 ). The primary outcome for Aim 2 is the Gross Motor Function Measure (GMFM), a measure of real-world locomotor capacity (50 (link)). Subscore dimension C (crawling and kneeling) will be used to determine change in prone locomotor function after prone training, and subscore dimension E (walking, running, jumping) will be used to determine change in upright locomotor function after upright training. See Table 1 for a summary of predictor, moderator and locomotor skill measures.