Subjects
Twenty-three participants with stroke were recruited and enrolled in this study. They had to be >2 weeks after their first clinical stroke (without an upper limit on post-stroke weeks). Participants with stroke were recruited on a patient-by-patient basis among the patients receiving an inpatient neurological rehabilitation at the Kliniken Schmieder Allensbach, Germany. Exclusion criteria for the participants with stroke were inability to detect any manually applied large passive finger movements. Other exclusion criteria regarding somatosensory deficits were not defined in this exploratory study to have a heterogeneous sample population with potentially a wide range of levels of proprioceptive deficits. Additional exclusion criteria were severe hand edema, high muscle tone—particularly in the flexor digitorum superficialis muscle and the flexor digitorum profundus muscle—evaluated with the Modified Ashworth Scale [
68], or pain preventing the use of the robotic assessment tool, severe cognitive impairment, aphasia, and neglect. If participants with stroke had difficulties with understanding the goal of the study and the instructions, the Montreal Cognitive Assessment (MoCA) [
69] was performed as a screening cognitive exam. In case of a value below 26 points on the MoCA scale, the participant was excluded. The presence of neglect symptoms was assessed by clinical observation. In case neglect was suspected, the Bells Test [
70] was performed. Twenty-nine neurologically intact control (NIC) subjects within the same age range served as a control group. Only self-reported right handed subjects were included to avoid a handedness confound when comparing proprioceptive performance between the dominant and non-dominant hands in participants with LHS and RHS. Handedness was assessed with the Edinburgh Handedness Inventory (left handed: score <−40, right handed: score >40, ambidextrous otherwise) [
71]. Participants with stroke were asked to evaluate their post-stroke handedness retrospectively. The study was approved by the institutional ethics committee of the University of Konstanz. All subjects gave signed, written informed consent in accordance with the Declaration of Helsinki before participating in the experiment.
Clinical assessments
Proprioceptive function was assessed based on the up-down test described by [
22]. The distal phalanx of the index finger was moved up or down, 5 times each, in random order. Participants with stroke reported the direction of the movement verbally in absence of vision of the tested finger. The final score (0–10) consisted of the number or correctly identified movement directions.
In addition, other somatosensory modalities were tested. Topesthesia (localization of touch) was tested by manually stroking the dorsal side of the fingers (2x per finger, random order). The outcome measure was the number of correctly identified fingers (0–10). Von Frey hairs (OptiHair
2, MARSTOCKnervtest, Schriesheim, Germany) were used to assess the absolute tactile perception threshold on the fingertip of the index finger. The score was computed by taking the geometric mean of the reverse values (5 suprathreshold and 5 subthreshold) of the descending staircase according to [
81], on a scale from 1 (0.25 mN) to 12 (512 mN). Pallesthesia (sensation of mechanical vibration) was assessed using a 64 Hz, graduated Rydel-Seiffer tuning fork (Martin, Tuttlingen, Germany) [
82] on the MCP joint of the index finger. The sensibility was scored from 0–8 in steps of one with 0 corresponding to no sensation at all. Stereognosis (ability to recognize objects by using only tactile information) was assessed with the subscale of the Nottingham Sensory Assessment [
22]. The outcome was the number of correctly identified objects (0–10).
Attention and working memory were assessed using the backward recitation condition of the Digit Span subtest of the Wechsler Adult Intelligence Scale
Ⓡ
- Third Edition (WAIS
Ⓡ
-III) [
83], where participants with stroke were asked to recite an auditorily presented series of digits backwards. Two trials with random numbers were consecutively performed for each digit span (2–6 numbers, in increasing order). The total score consisted of the total number of correctly recited digit spans ranging from 0–12.
Data analysis
To evaluate the test-retest reliability (in participants with stroke), the intraclass correlation coefficient
ICC(2,1) (two-way layout with random effects for absolute agreement) [
84], as well as its 95% confidence interval (CI), standard error of measurement
SEM, and smallest real difference
SRD (sometimes referred to as minimal detectable change
MDC) were computed according to [
85] and [
86]. Systematic bias was analyzed by calculating the mean difference
\(\bar {d}\) between the two test occasions and its 95% CI, and by visualization in a Bland-Altman plot [
85]. The reliability analysis was performed separately for the impaired and unimpaired hand.
To compare the outcome measures of NIC subjects to participants with stroke and to create models of neurologically intact performance, the PEST sequences of the NIC subjects were truncated to 60 trials to be of same length as for the participants with stroke. Furthermore, outliers in the NIC group were identified according to Tukey’s rule and excluded from all statistical analysis.
Outcome measures from the robotic assessment were compared between the left and right hand of NIC subjects (paired test), between the impaired hand of the participants with stroke and the corresponding hand in NIC subjects (two unpaired tests for LHS and RHS), and between the unimpaired hand of the participants with stroke and the corresponding hand in NIC subjects (two unpaired tests for LHS and RHS). For participants with stroke, the average of both test and retest outcomes was used. Two-sample t-tests or Wilcoxon rank sum tests, respectively paired-sample t-tests or paired Wilcoxon signed rank tests, were conducted depending on whether data (or their differences for paired testing) were normally distributed or not. Normality was tested with the Shapiro–Wilk or the Shapiro–Francia test, depending on the kurtosis. To correct for multiple comparisons, a Šidák-correction was used. To compare the effect of a LHS versus a RHS on proprioceptive function of the impaired hand (as well as unimpaired hand), their differences to NIC baseline of the corresponding hand were compared in an unpaired test. As NIC baseline the median was used, as the distributions of the left and right DL in NIC subjects were not both normally distributed. Again, the test-retest average was used for the participants with stroke.
Subjects in this study were not age-matched on an individual basis but on a group level by including subjects within the same age range. In return, the sample size of NIC subjects was chosen to be larger, which should improve the estimated distribution of DLs for the same age range. To model neurologically intact performance in elderly, log–normal (semi-infinite positive support) probability density functions were fitted on the DL data of NIC subjects for the right (dominant) and left (non-dominant) hand separately. The 95th percentile was used to characterize impairment in participants with stroke. The CI for the 95th percentile cutoff were calculated by bootstrapping.
The trial duration of the robotic assessment was compared between impaired and unimpaired hands of participants with stroke (averaged test-retest, paired-sample t-test or paired Wilcoxon signed rank test, depending on normality of the paired differences), and between NIC subjects (mean of both hands for each subject) and the impaired, respectively unimpaired hand of participants with stroke (averaged test-retest, two-sample t-test or Wilcoxon rank sum test, depending on normality of the two distributions). The Šidák-correction method was used to correct for multiple comparisons.
Spearman’s rank-order correlations were calculated for the impaired hand between the DL (test-retest average) and the outcome measures of the clinical scales (i.e., up-down, localization, Von Frey hair, vibration, stereognosis, working memory). As the up-down test also measures proprioception, the correlation with the robotic assessment could be regarded as a test for concurrent validity. The clinical scores for the impaired hand and the working memory test were also compared between the participants with RHS and LHS with two-sample t-tests and Wilcoxon rank sum tests, respectively.
The influence of sex on the proprioceptive outcome measures was tested by separately comparing the robotic outcome measures for the dominant as well as the non-dominant hand in male versus female NIC subjects, and for the impaired as well as the unimpaired hand in participants with stroke (two-sample t-test or Wilcoxon rank sum test, depending on normality of the distributions).
Significance levels were set to α=0.05. Probability values p<0.05 and p<0.01 are marked as * and **. Descriptive statistics are reported as mean ± SD, unless otherwise stated. For non-parametric statistics the median and interquartile range (IQR) was reported. All statistical analyses were performed in MATLAB R2014a (MathWorks, Natick, MA, USA).