Ankle proprioception in children with cerebral palsy

2.1 Participants

A power analysis using an effect size estimated from between group differences observed in Zarkou et al. informed the target sample size of children with CP [14]. For power = 0.8, alpha = 0.05, and Cohen’s d ≈ 0.96, a minimum of 15 participants per group were necessary to detect a statistically significant difference in ankle position sense acuity. Fifteen children with CP (mean [M] age: 11 years 10 months; standard deviation [SD]:±2 years 10 months; range: 6 years 8 months –15 years 9 months; 6 males, 9 females) and 58 TD children (M age: 12 years 3 months; SD:±3 years 2 months; range: 7 –17 years; 34 males, 24 females) were able to follow the instructions and complete the study (Table 1). All children with CP were recruited from patients undergoing, or who had recently undergone, a clinical gait analysis at Gillette Children’s. All but four completed the proprioception test immediately after a 2.5-hour clinical gait analysis. Inclusion criteria were 1) a diagnosis of CP, 2) Gross Motor Function Classification System (GMFCS) level I-III, 3) age six to < 18 years, 4) able to comprehend English, and 5) expected to be able to follow instructions. Ambulatory individuals were enrolled as part of a larger project that required participants to be able to ambulate and perform other standing balance tasks. Exclusion criteria were any surgery in the past nine months or botulinum toxin injection in the past three months. One participant was receiving intrathecal baclofen. All TD children were recruited and tested at the 2019 Minnesota State Fair. Inclusion criteria were 1) no reported history of central or peripheral nervous system disorder, 2) no current injury to the lower limbs, and 3) no exposure to chemotherapy which could have affected somatosensory and motor function. Before testing, TD children completed the footedness questionnaire to determine the dominant foot to be tested [16]. Appropriate written parental consent and child assent were obtained prior to data collection. The studies were approved by the University of Minnesota Institutional Review Board.

Data were collected using the manual Ankle Proprioceptive Acuity System for all participants (Fig. 1A). Feasibility of this system for measuring human ankle position sense acuity had been established previously [17]. Moreover, intra- and inter-rater reliability concerns were negligible because the applied psychophysical method was not subject to researcher bias. The researcher did not rate a perceiver’s performance but only rotated the joint to the position indicated by the psi-marginal adaptive algorithm. With respect to test-retest reliability, the only source of variability was the inherent variability of the responder’s perception of ankle position. Unpublished data from eight healthy adults tested on three consecutive days showed very low test-retest variability (standard error of measurement = 0.09°).

OPEN IN VIEWER

Fig. 1

A. Front and side view of the manual Ankle Proprioceptive Acuity System. Rotating the handle by the researcher rotates the ankle. Desired degree of rotation can be set by mechanical stops in the semicircular pegboard at 0.1° increments. The system components can be adjusted to the leg anthropometrics of the participant, so that the approximate ankle joint axis aligns with the axis of rotation of the device. B. Example of the time course of a single trial. All trials started at the neutral 45° position. Reference position was always at 15° plantarflexion. Here, the researcher rotates the foot to the reference position, then returns the footrest to the initial position before rotating it to the comparison position (10° plantarflexion in this example). Thus, the stimulus difference is 5°. The participant indicates which experienced position (1^st or 2^nd) was closer to the floor. C. Example of a derived psychometric function. The just noticeable difference threshold corresponds to the difference between reference and comparison position at the 75% correct response rate. The uncertainty area is defined as the distance between the stimulus size differences at the 60^th and 90^th percentiles.

The system allowed for manual plantarflexion/dorsiflexion of the tested ankle to distinct ankle positions with a pegboard position resolution of 0.1°. Angular position and velocity were recorded by a US Digital H6 Optical Encoder (resolution: 0.036°) embedded in the device and aligned with the participant’s ankle joint axis (sampling frequency 100 Hz).

Participants were barefoot and seated on a chair to perform the ankle position discrimination test on the more-involved (CP) or dominant ankle (TD). The more-involved side was determined by clinical diagnosis or a comprehensive exam (e.g., contracture, spasticity, weakness, selective motor control) by the physical therapist during the gait analysis. The footrest height and heel rest position were adjusted to align the ankle center, approximated as the lateral malleolus, with the device axis of rotation. The tested foot, stabilized by a strap, sat on the footrest with the participant’s ankle at approximately 90° relative to the shank. Participants wore vision occluding glasses to remove possible visual cues of ankle position during testing.

One of three researchers slowly plantarflexed the participant’s foot from an initial neutral position (0° plantarflexion) to a reference (15° plantarflexion) or comparison (0-15° plantarflexion) position. The participant’s foot was held for approximately two seconds, then moved back to the initial neutral position before moving to the other reference or comparison position (Fig. 1B). Participants were asked to verbally indicate in which of the two positions (first or second) their toes were closer to the floor. They were told to provide their best guess if they were unsure. Also, they were allowed to repeat a trial if they were not paying attention. Before the testing, there were at least three practice trials with or without vision occluding glasses to help participants acclimate to the device and testing procedure. The practice trials began with a large difference between the reference and comparison position (approximately 10°), to ensure that participants were able to discriminate and to understand the testing scheme. During actual testing, a participant’s verbal response (incorrect/correct) in the previous trial was used as input for an adaptive psi-marginal algorithm to determine the comparison position for the next trial within the allowable 15° stimulus range [18]. A total of 25 trials were performed. After 10 trials or when participants requested, a break was given. Instructions were reiterated when necessary. The complete protocol including practice and 25 test trials took 15-30 minutes to complete.

2.4 Psi-marginal adaptive algorithm

The psi-marginal adaptive algorithm was used to update which comparison ankle position to present to the participant for each test trial [18]. This algorithm effectively decreased the number of trials (and thus time) necessary to determine the smallest detectable stimulus difference that a participant could perceive. The four parameters for this algorithm include the threshold (alpha), slope (beta), upper asymptote (i.e., lapse rate; lambda), and lower asymptote (i.e., guess rate; gamma). The threshold was of primary interest and was the stimulus intensity (difference between the reference and comparison ankle positions, in degrees) at which a participant would discriminate two different ankle positions at 75% accuracy. The range in which each participant’s threshold was assumed to fall was 0-15° of plantarflexion, with an increment of 0.1°. The slope measured the variability in perceptual judgments during an assessment. The range in which each participant’s slope was assumed to fall was -1.2 to 1.2. Lapse rate represented the proportion of trials the participant incorrectly answered because of inattentiveness for stimuli that the participant was truly capable of differentiating when vigilant. The range in which each participant’s lapse rate was assumed to fall was 0 to 0.1. Guess rate for this two-alternative forced choice task was 0.5, meaning that participants could have guessed correctly 50% of the time. Wait time was set to four, which represented the number of trials after the maximum stimulus intensity was presented (15° in this study) in which lapse rate was assumed fixed (0.1 in this study). Thus, if a wrong answer occurred during this wait time, lapse rate was not updated in the posterior distribution but rather stayed fixed.

2.5 Psychometric function fitting

A logistic Weibull function (i.e., Gumbel function) was fitted to the stimulus difference-response data (i.e., the difference in the reference and comparison ankle angular positions and the incorrect/correct verbal response data for each participant) [19]. Based on the fitted function (Fig. 1C), two components of ankle position sense acuity were calculated: bias (systematic error) and precision (random error). Bias was measured as the JND threshold, the stimulus difference between the reference and comparison angular positions (in degrees) at the 75% probability of correct response of the function [15, 20]. Smaller JND thresholds (or less bias) represented higher ankle position sense acuity, which implied that the participant could discriminate between smaller differences in ankle position. Precision reflected how consistently a participant responded across all trials, represented by the UA of the fitted function. UA was calculated as the range of the stimuli (in degrees) between the 60% and 90% probabilities of a correct response [20]. A smaller UA value represented a higher ankle position sense precision, that is, the participant was more certain in their responses. The function fit allowed a range of UA≤25°.

2.6 Statistical analysis

The statistical analysis was based on custom-written code in R (version 4.1.2). Shapiro-Wilk tests were performed to determine the normality of the outcome variables. Given that all variables were non-normal, subsequent non-parametric Wilcoxon rank sum tests were applied to determine group differences (TD vs. CP) for JND threshold and UA. Statistical significance was defined as p < 0.05.

4.1 Study limitations

There were several limitations to this study that need to be considered. First, this proof-of-concept study, which focused on high functioning children at GMFCS levels I-III, only examined a relatively small group of children with CP. Currently, there is limited or no data on children in GMFCS levels IV-V. Yet, it would be imperative to obtain such data to discern how proprioceptive dysfunction involving the ankle contributes to deficits in gross motor ability. Second, like all psychophysical testing procedures, the method applied here required that the examinee had the cognitive ability to comprehend the task and stay focused and attentive during testing. This limited the applicability of the test to school-age children and to the approximately half of the CP population without an intellectual disability [32]. Variability of maintaining attention, especially in younger children, was evident in this study. Only 29% (2/7) of the children with CP younger than nine years old successfully completed the testing, compared to 62% (13/21) of those 9-15 years old. Other researchers who have performed sensory testing with children also noted that younger children are less likely to be able to fully participate and yield valid data [7]. To ensure that psychophysical testing of children with CP yields meaningful data, researcher need to create a test environment that allows for breaks and keeps the children engaged throughout testing. Third, the dominant ankle in the TD group was compared to the more involved/non-dominant ankle in the CP group. This approach was chosen since no significant differences between dominant and non-dominant ankle proprioception have been reported in TD individuals [33] and because the more-involved side has the most impairment in CP. Future studies should explore the differences between sides of asymmetrically involved individuals and the impact on gross motor function. Fourth, siblings were allowed to participate, so the TD group may not all be independent observations. Finally, high lapse rates can spuriously increase JND thresholds [19, 35]. While the participants’ lapse rates in this study were not objectively measured, it was assumed that they ranged from 0-30% as others have observed [34, 35]. If true lapse rate (denoted as lambda in the psychophysical function) was greater than the estimated lambda from the adaptive psi-marginal algorithm (restricted to≤10%), JND thresholds presented here would be overestimated, indicating poorer acuity than their true ability [19]. However, there is a practical trade-off between more precise estimates of JND threshold and UA by collecting more trials and increased likelihood of attentional lapses with longer testing.