Motion Characteristics of Subclinical Tremors in Parkinson’s Disease and Healthy Subjects

The Parkinson’s disease tremor characteristics reported previously are not applicable to the full spectrum of severity. The characteristics of high- and low-amplitude tremors differ in signal regularity and frequency dispersion, which indicates that characterization should be studied in separate severity. The subclinical tremor of Parkinson’s disease is close to physiological tremor, yet their distinctive features are still undetermined. This study aims to determine joint motion characteristics that are unique to subclinical Parkinson’s disease tremors. The tremors were characterized by four hand–arm motions based on displacement and peak frequencies. The rest and postural tremors of 63 patients and 62 normal subjects were measured with inertial sensors. The baseline was established from normal tremors, and the joint motions were compared within and between the two subject groups. Displacement analysis shows that pronation–supination and wrist abduction–adduction are the most and least predominant tremor motions respectively, for both Parkinson’s disease and normal tremors. However, the subclinical Parkinson’s disease has signicant greater in amplitude and peak frequency in specic predominant motions as compared to normal tremor. The exion-extension of normal postural tremor increases in frequency from proximal to distal segment, which is explainable by mechanical oscillation. This characteristic is also observed in patients but with amplication in wrist and elbow joints. The contributed distinctive characteristics of subclinical tremors provide clues on the physiological manifestation that is a result of the neuromuscular mechanism of Parkinson’s disease.


Introduction
Characteristics of Parkinson's disease (PD) tremors in general have been studied extensively. The rest and postural tremors of PD have been reported to be of 3-12Hz, 1,2 with peak found not only at fundamental frequencies but also at harmonics. 2 Finger exion (pill-rolling motion), wrist exion and extension, pronation, and supination were also observed in PD rest tremor. 3 These characterizations were made on tremors that were visible, and are not be applicable to the full spectrum of severity.
One of the few researches on PD of low and high amplitudes and healthy controls reported the tremor characteristics based on the signal regularity or predictability of future value in time series. The regularity was quanti ed by approximate entropy, the amplitude of which indicates the randomness of a signal (Approximate entropy ranges from 0 to 2; the value 0 indicates accurate short-and long-term predictions of future value, as seen in a sine wave; the value 2 indicates that a signal is highly randomized such as white Gaussian noise). 4 The study shows that the approximate entropy gradually decreases across the subject group of control, least affected limb of the PD patient, and most affected limb of the PD patient.
In other words, low-amplitude tremor has a less regular signal.
Furthermore, the proportion of acceleration power, which is a ratio of power in individual frequency bin to the total power from 1Hz to 30Hz in a power spectral analysis, in the least affected limb of the PD patient is less than the most affected limb of the PD patient at around 9Hz, and it is the other way around at 16-30Hz. 5 The difference in the characteristics of high and low amplitudes suggests that PD tremors should be studied at a separate severity level.
The subclinical PD tremor or tremor that is not easily detectable in PD, is known to be close to physiological tremor. 5 Based on previous studies, the amplitude of displacement and/or the peak frequency alone was insu cient to establish a signi cant difference between the two types of tremors. 1,5 Few studies show that time-and frequency-varying properties are signi cantly different in subclinical tremors of PD and controls. 1,6 However, the distinctive features of these two types of tremors are still undetermined. Apart from this, the clinical impressions of certain joint motions to characterize PD tremors are found, 3 but no relevant evidence from the measurement on subclinical PD tremor is available.
Quantifying tremor in joint motion and different segmental location provides more detailed phenomenological data on the unobservable tremors. In this observational study, we are interested to nd the joint motion characteristics that are unique to PD with no clinical tremor signs. For the rst time, the tremor is characterized in four hand-arm joint motions by using the displacement and peak frequency.
The key clinical ndings from the study were the predominance of certain joint motions and distinctive tremor motion characteristics of PD with subclinical tremor.
Furthermore, physiological tremor has been used to explain the origin of pathological tremor, 7 which suggests the importance of the tremor of healthy subjects. Thus, in this work, the tremor data from healthy subjects were presented and tremor motion within and between normal and PD subjects were compared. This study provides the physiological presentation that may help explaining the neuromuscular mechanism in further study.

Clinical characteristics of subjects
In this study, the mean ages of the 63 PD and 62 normal subjects participated are 70 (standard deviation, SD = 7.8) and 55 (SD = 10.8), respectively. The durations from the last intake of medication to the rst measurement differ among subjects. An estimated medication wear-off period of three hours was used as a reference to characterize the patients recruited. More subjects took the medicine for three or more hours (n = 45; 71.4%). Five (7.9%) subjects who could not report that duration were categorized as unknown for that criterion, and one subject was not on medication. The generalization and limitation of the study are presented in the supplementary material.
Within-group tremor motion comparison Different motions within each type of tremor were compared using RMS ∆θ joint and peak frequency. The η 2 values of 0.09 to 0.77 were obtained for all the cases with signi cant difference, which indicates that the effect of all the differences ranges from medium to large based on the interpretation guidelines provided by Cohen (1988) (small effect η 2 = 0.01; medium effect η 2 = 0.06; large effect η 2 = 0.14). 13 Between-group tremor motion comparison PD as well as normal tremors were rated below 0.5 based on predicted rating. Nevertheless, the Kruskal-Wallis test on RMS ∆θ joint shows that the PD tremor in every motion of three tested postures is signi cantly different from the normal tremors. The median of RMS ∆θ joint values of the tremors is reported in Fig. 3A to C. The rankings of severity in the blue bars show that the PD tremors have the following decreasing order of severity in rest and outstretching actions: EPS, EFE, WFE, and WAA.
In controls, the tremor severity rank for all tremor motions is the same except for the wrist tremor motions in resting posture, which is not signi cant difference from each other. Thus, the WAA and WFE have the same rank (refer to Fig. 3A).
Further analyses of PD tremor show that the median values of relative severity of EPS over EFE (EPS minus EFE) and of EPS over WAA (EPS minus WAA) are 1.75 and 1.5 times greater, respectively, than (p < 0.001) the median values of normal tremor during resting condition (refer to Fig. 3A). Similarly, in outstretching and wing postures, the median values of relative severity of EPS over WAA are 1.82 and 1.18 times greater, respectively, than (p < 0.005) those of normal tremor.
The analyses of peak frequency (Fig. 4) show that the peak frequency of the PD tremor is close to that of the normal tremor in every joint motion of all actions, except in outstretching (refer Fig. 4B) posture, and the PD tremor has relatively higher frequency compared to normal tremor for EFE and WFE. The peak frequency of the rest of the cases ranges from 4.1Hz (95% CI = 3.8Hz, 4.8Hz) to 5.7Hz (95% CI = 4.9Hz, 6.8Hz) in PD tremor and 4.0Hz (95% CI = 3.8Hz, 4.1Hz) to 5.3Hz (95% CI = 4.2Hz, 6.1Hz) in normal tremor.
The effects of most of the between-group comparisons with signi cant difference are medium to large based on the η 2 of 0.06 to 0.22. The only cases with small effects are the between-group comparison based on EPS-WAA in terms of RMS ∆θ joint for wing action (η 2 = 0.05) and WFE in terms of peak frequency for outstretching (η 2 = 0.05). The values of amplitudes in RMS ∆θ joint and peak frequency and the corresponding η 2 and p values are documented in the supplementary material.

Important ndings
Though all the tremors had no clinical signs, the characteristics that are unique to PD and normal tremors are found in the analysis (see Table 1). In between-group comparison especially, PD tremors are signi cantly different from normal tremors. The median values of the rest and postural conditions of baseline and PD tremors having no clinical sign based on predicted rating are reported for the rst time.
The severity ranking of the PD and normal tremors in increasing order is WAA, WFE, EFE, and EPS. This order is found in all actions except for the WFE and EFE in wing posture of PD tremors and the wrist motions in rest condition of normal tremor. The rank order indicates that the tremor about the elbow has larger amplitude compared to tremor about the wrist joint in most of the tremor cases.
PD and normal tremor share the common characteristics of having EPS as the most predominant tremor motion and WAA as the least predominant tremor motion. Nevertheless, such predominance in PD tremor is signi cantly greater than that of normal tremor, as supported by a signi cantly larger EPS-WAA value in the former tremor. This suggests that the relative severity of the two motions in PD tremor is not a mere manifestation of physiological tremor. Previous studies reported that postural tremor in exion-extension of distal segment is greater than proximal one. 15,16 This agrees with the exion-extension of normal postural tremor in our ndings. Our study further reveals that it is distinctive in different motions for the same segments, for example WAA has lower peak frequency than WFE. Another interesting nding is that the pathologic condition alters the tremor such that the peak frequencies are ampli ed in EFE and WFE only. In short, EFE and WFE in outstretching posture the only motions that give distinct difference between PD and normal tremors based on both amplitude and frequency. The ndings of rank order and distinct peak frequency of PD tremor in speci c motion are new.

Tremor motion characteristics found in other studies
Two studies on PD tremor show that EPS is a dominant tremor motion. One study reported on the anecdotal impression that EPS was mainly observed in PD tremor, 17 which agrees with the ndings of our study. Another clinical observational assessment on the tremors of 50 PD patients further reveals that the rating in WAA is the lowest compared to WFE and EPS in outstretching (WAA rating = 0.05 ± 0.18; WFE rating = 0.17 ± 0.40; EPS rating = 0.15 ± 0.46) and wing postures (WAA rating = 0.01 ± 0.07; WFE rating = 0.17 ± 0.42; EPS rating = 0.12 ± 0.40). 18 Nevertheless, statistical analysis was not made to further clarify the signi cant difference of the severity of the motions compared.
The previous ndings on comparing the EPS versus WFE, however, do not exactly match the characteristics of PD found in our study. The reported relative severities of WFE and EPS differ depending on postures (WFE is most dominant in outstretching posture; EPS is most dominant in wing posture), 18 contrary to our ndings that tremor in EPS appears to be the most severe for all the tested actions. The possible reason for the different ndings is the inclusion of visible tremor in the previous study.
Nonetheless, further studies are required to understand the variation in the ndings.

PD tremor of low amplitude versus normal tremor of other studies
The research on comparing low-amplitude PD and normal physiological tremor is limited. The early comparison studies were done using RMS displacement, median frequency, 1 and peak frequency, 5 and on locations -namely, the nger 1 and hand. 5 Though some PD patients recruited in these studies have visible tremor, 1,5 the analyses show that the two types of tremors are not distinguishable, particularly if only amplitude, of either displacement or frequency, is used for comparison. This highlights the strengths of our study, which is able to nd distinctive features of the two tremor types that are all of rating less than 0.5. The probable reason is that the system developed provides more degrees of freedom and more speci c joint motions and locations for analysis. This leads to more comprehensive tremor characterization even with only amplitude, rather than its derivatives. However, it is worth appreciating the studies that investigate the transient characteristics 1 and other derivatives, 6 as they have also provided other distinguishing features to the two tremor groups.
Previously, the frequencies of pathological tremor (i.e., PD and ET) were reported to be indistinguishable from physiological tremor. 6 A study that shows no signi cant difference in comparing tremors of controls and PD of mild to moderate amplitudes using frequency at peak power 5 also supports that frequency is di cult for making the differentiation. Most of the results in our study match with its ndings, but two special cases that are able to differentiate subclinical PD and normal tremors are found: WFE and EFE of outstretching posture. The study of comparing frequency at different tremor motions is new, and no previous clinical observation data are available for comparison.

Clinical implication
Based on the previous studies on the effect of ventrolateral thalamotomy on Parkinson's disease patients, the reduction of tremor in the directions captured by sensors, particularly in EPS was observed. 19 Based on our study, subclinical PD tremor is not unidirectional. Rather, it is a symphony of joint motions that have unique predominance of amplitude and frequency in speci c direction. To the best knowledge of the authors, the neuromuscular mechanism that causes such combination of tremor motions has yet been investigated. The ndings of this work are hoped to lead to more insights of the central oscillation and involuntary muscle actuation that is responsible for the subclinical PD tremor.
In conclusion, we have contributed the evidence of the amplitude-and frequency-based predominance of tremor in individual motions and the joint locations at which tremor is captured. Besides, the ability to draw a line between the tremors of subclinical PD and controls based only on amplitudes suggests that joint motions provide more comprehensive characterization. Thus, characterizations with the derivatives, typically the temporal and frequency change in tremor motions, are worthy of further study. This work has provided a clue to the physiological manifestation that is a result of neuromuscular mechanism of subclinical PD tremor.

Study Settings and Participants
This is a cross-sectional study that centers around the characterization of tremor based on hand-arm motion. The four hand-arm motions studied are wrist exion-extension (WFE), wrist abductionadduction (WAA), elbow pronation-supination (EPS), and elbow exion-extension (EFE). PD patients and controls (i.e., normal subjects) are the two subject groups involved. The study encompasses (i) the establishment of baseline values using the normal subject tremor readings, (ii) within-group tremor motion comparisons, and (iii) between-group comparisons.
With the approval of the Medical Research Ethics Committee, Secretariat of National Institutes of Health, Malaysia, a study (protocol no. NMRR-14-1694-21740 (IIR)) of four months in duration was carried out mainly in the Neurology Clinic of Penang General Hospital. The research methods carried out were in accordance with the National Institution of Health guidelines. Sixty-three PD patients attending walk-in and appointment clinics and 62 normal subjects were recruited following written informed consent from either the patients or the legal guardians of the patients. The total number of participation and missing data of each case of study are presented in the supplementary material.
All subjects were recruited based on a number of inclusion criteria. All subjects were adults aged 40 years old or above. The PD patients were diagnosed of idiopathic PD based on the United Kingdom Parkinson's Disease Society Brain Bank clinical diagnostic criteria. 8 They were screened to include those with no signi cant clinical tremor by using a predicted tremor rating of below 0.5 as one of the inclusion criteria. The tremor ratings were predicted using a regression model that relates the observational ratings from doctors and the readings of the biomechanical system in a previous study. 9 The predicted rating in assessing tremor occurring during rest and outstretching postures is equivalent to the MDS-Uni ed Parkinson's Disease Rating Scale (MDS-UPDRS), and the predicted rating in assessing tremor during wing posture is equivalent to the Washington Height-Inwood Genetic Study of Essential Tremor (WHIGET) rating scale (wTRS). A rating of 1 indicates the minimum tremor severity that is observable based on both the MDS-UPDRS and wTRS, and setting a predicted rating of 0.5 as a threshold, is a more conservative means to categorize a tremor as a non-signi cant clinical sign.
The key criterion to include the normal subjects was the absence of tremor-related disease or illness. The exclusion criterion for both groups was the intake of substances or drugs that induce or suppress tremor, with the exception of the intake of PD medication for the patients.

Procedure
Before the measurement, the tremor-related illness history, disease duration, and time since last dose of the tremor-suppressing medicine were recorded. Each subject was then asked to count numbers in decremental order with two per interval and perform the speci c posture for an allocated time. The upper limb resting and outstretching postures were performed according to the protocols in the MDS-UPDRS upon attainment of permission from the International Parkinson and Movement Disorder Society, and the wing posture was done according to the protocol of the wTRS. 10 The following actions were performed: i. The upper limb was rested on the arm rest for 15s.
ii. The upper limb was outstretched in front of the chest for 15s.
iii. The arm was held in wing position for 15s.
Since the patients were not recruited upon appointment, some of them took tremor-suppressing medicine, and they had different durations between measurement and last dose of medication (refer to the clinical characteristics in the results for the data).

Tremor measurement
Four hand-arm motions were quanti ed with the use of the Attitude and Heading Reference System (AHRS). The AHRS of model SBG IG-500A (SBG Systems, Rueil-Malmaison, France) consists of triaxial magnetometers, triaxial accelerometers, and triaxial gyroscopes.
In order to measure the relative motion of the wrist and elbow joints, one AHRS was a xed on the hand, the lower arm, and the upper arm. The quaternion in each AHRS was processed to attain joint angle. 9 Subsequently, a fourth-order Butterworth bandpass ltering with a passband of 3-30Hz was performed to limit the signals to contain only physiological and pathological tremors, which were previously found to be within this range. 11 The resulting parameter of the ltering is termed joint angular displacement, ∆θ joint , which is essentially the displacement about the joint (in terms of degree, ˚) that includes mainly tremulous information. The time series ∆θ joint in four tremor motions were computed for root mean square (RMS), and the serial data were also subjected to fast Fourier transform spectral computation using a Hanning window. The RMS and peak frequency of ∆θ joint were the key characterization parameters of the tremor motions. The measurements were performed by the research assistants, and the data acquisition and data processing were done automatically in the LabVIEW™ software (National Instruments Corporation, Austin, Texas).
The measurement system was validated by comparing its readings with those of the tremor simulator in the laboratory. The coe cient of determination, R 2 of the linear regression relating the RMS of angular displacement of the two systems, is 1.0000 (p < 0.001). 9 In measuring PD tremors, during resting, outstretching, and wing actions for clinical validation, the regression analysis shows that the reading of the measurement system can explain more than 80% of the variability of the doctor's observational rating (R 2 > 0.80) in our previous study. 9 Statistical analysis Non-parametric statistics were used for the analyses because the data were not normally distributed. The analyses were carried out using IBM SPSS Statistics for Windows version 23.0 (IBM Corp., Armonk, NY, USA). In within-group comparison, the Wilcoxon-signed rank test was used to test the statistical signi cance of the readings in four tremor motions and rank the severity of the tremor. When performing multiple comparisons, some statistical tests may result in p values of less than 0.05 by chance, so Holm's sequential Bonferroni correction is one way to resolve the problem by adjusting the p values. Six sets of within-group comparisons are possible. The rst to sixth most signi cant p values must be less than 0.008, 0.010, 0.013, 0.017, 0.025, and 0.050, respectively, to be considered to have signi cant difference. The method to obtain all the signi cant p values is elaborated in the supplementary material.
The Kruskal-Wallis test was used to compare the tremors between normal and PD groups. The tremor characteristics compared between the two groups are the readings of individual tremor motions and the pair-motion difference (i.e., the difference between the most severe and other individual motions). The three pair-motion differences identi ed were the EPS-EFE, EPS-WFE, and EPS-WAA after understanding that EPS is the most dominant motion. The effect size in eta-squared, η 2 is calculated for the evaluation of the size of the difference of the tremor characteristics within the same subject group and between the two groups.
The 95% con dence interval (CI) of the median of all parameters was estimated by the non-parametric bootstrap method because the data were not normally distributed. 12 In this method, 10,000 bootstrap samples were generated from the readings of each case of measurement (i.e., in resting and posture maintaining conditions). Each bootstrap sample has the same sample size as the original sample, and all the tremor parameters (peak frequency and RMS of ∆θ joint in all motions) were made sure to be associated with the same sets of subjects. The 10,000 median values were then computed.
Declarations Figure 3 Median of RMS ∆θjoint of subclinical PD and normal tremors for (A) resting, (B) outstretching and (C) wing postures. The arrows indicate the relative severity of the pair motions. The severity rank is marked below each bar (higher rank indicates greater severity). The * marked above and below the graphs indicate the signi cant difference of the pair and individual motions respectively. The signi cant difference is reported at * p < 0.05, ** p < 0.001 and *** p < 0.0001.