Abstract
Background:
Prior studies have relied on conventional observer-based severity ratings such as the Unified Huntington’s Disease Rating Scale (UHDRS) to identify early motor markers of decline in Huntington’s disease (HD).
Objective:
The present study examined the predictive utility of graphomotor measures handwriting and drawing movements.
Methods:
Seventeen gene-positive premanifest HD subjects underwent comprehensive clinical, cognitive, motor, and graphomotor assessments at baseline and at follow-up intervals ranging from 9–36 months. Baseline graphomotor assessments were subjected to linear multiple regression procedures to identify factors associated with change on the comprehensive UHDRS index.
Results:
Subjects were followed for an average of 21.2 months. Three multivariate regression models based on graphomotor variables derived from a complex loop task, a maximum speed circle drawing task and a combined task returned adjusted R2 coefficients of 0.76, 0.71, and 0.80 respectively accounting for a significant portion of the variability in cUHDRS change score. The best-fit model based on the combined tasks indicated that greater decline on the cUHDRS was associated with increased pen movement dysfluency and stroke-stroke variability at baseline.
Conclusion:
Performance on multiple measures of graphomotor dysfluency assessed during the premanifest or prodromal stage in at-risk HD individuals was associated with decline on a multidimensional index of HD morbidity preceding an HD diagnosis.
INTRODUCTION
Huntington’s disease (HD) is an inherited progressively disabling disorder that leads to disturbances in motor function, cognition, and behavioral control. The time course from onset of symptoms to death ranges from 10–20 years [1]. Over 25% of the cases present with chorea as their first symptom. Identifying behavioral, biologic, or neuroimaging predictors of outcome in HD has been a priority for large international research groups including PREDICT-HD [2, 3]. Cytosine-adenine-guanine (CAG) repeat length and age of onset of first signs of disease are considered useful markers of disease burden [4, 5].
Over the past several years members of the PREDICT-HD consortium and others have identified several clinical and imaging predictors of motor decline and/or onset of manifest HD (see Paulsen [6] for comprehensive review). For example, Harrington and colleagues [7] followed gene positive prodromal individuals assessed along six cognitive factors to identify predictors of disease progression. Performance on tasks within the motor planning/speed and sensory-perceptual processing domains predicted time to clinical diagnosis after controlling for baseline CAG-age product (CAP) and motor symptoms. In a 36-month longitudinal study of gene-positive premanifest HD cases, Tabrizi et al. [8] reported worsening of performance on several motor and cognitive tasks among cases who ultimately received a diagnosis of HD. Quantitative motor and cognitive measures were also associated with functional decline in early manifest HD cases.
In what is perhaps the largest longitudinal study of premanifest HD cases, Paulsen and colleagues [9, 10] used survival analysis to identify three phenotypic domains to predict decline in motor function in over 1000 HD cases. Statistically significant predictors of motor diagnosis included total motor score of the UHDRS, performance on the Stroop-Word test (SWT), symbol-digit modalities test (SDMT) score, total functional capacity scale rating, age, and CAG. Using principal component analyses, Langbehn et al. [11] reported significant correlations between UHDRS motor and cognitive scores and baseline SDMT and SWT. Ghazaleh and colleagues [12] followed 1,608 HD cases and identified a non-linear model consisting of baseline CAP, CAG repeats, tetrabenazine use, antipsychotics use, significant cognitive impairment or dementia, history of apathy, and age at diagnosis accounting for 41% of the variability in outcome composite and total motor scores of the UHDRS. Many of these studies are remarkable in their ability to enroll and follow large samples of premanifest HD cases. Overall, prior studies suggest that early prediction of HD motor diagnosis in at-risk patients can be enhanced with early assessment of relevant motor and cognitive tasks.
With a focus on quantitative measures of motor function in HD, our group recently examined whether subtle graphomotor (or handwriting) abnormalities are present prior to clinically manifest chorea in HD [13]. Thirty-eight symptomatic HD, 30 gene-positive premanifest (PM), and 25 normal healthy control subjects completed graphomotor tasks consisting of circles, loops, sentences, and spirals with a non-inking pen on a digitizing tablet. Multiple kinematics and pressure variables were measured along with the cognitive and motor status of each participant. HD subjects exhibited significantly longer and more variable stroke durations, decreased handwriting smoothness, and increased and more variable pen pressures compared with the healthy controls, findings consistent with prior work on handwriting movements in HD [14–16]. Furthermore, a multifactorial statistical model consisting of graphomotor variables differentiated gene-positive PM from gene-negative healthy subjects with 85% accuracy. These findings suggest that graphomotor measures may be useful as early predictors of decline in premanifest HD.
Motor assessments employed in prior HD prediction studies relied on conventional subjective observer-based rating scales such as the motor subscale of the UHDRS. While comprehensive, the scale is insensitive to subclinical motor abnormalities, rates severity of motor signs using a nonlinear ordinal scale and requires training and experience to produce reliable results. Quantitative instrumental measures generally overcome these limitations; however, their application in the clinical setting is often limited by lack of technical expertise and laboratory space to house the instruments. Wearable sensors such as wrist actigraphy or body accelerometers are sensitive to hyperkinetic movements [17, 18] while recording movements in a naturalistic setting. Similarly, graphomotor procedures are particularly sensitive to dopamine-mediated hyperkinesia and hypokinesia through analyses of pen stroke kinematics using familiar tasks such as handwriting and drawing [13, 19]. Based on prior large-scale studies showing that motor function is a predictor of disease progression in HD, we reasoned that a graphomotor procedure would be equally robust as a predictor of decline but would require far fewer cases because of their increased precision and sensitivity.
The present study examined the predictive utility of graphomotor measures handwriting and drawing movements in premanifest HD. Premanifest refers to cases early in their stage of disease, in which motor or cognitive symptoms are not yet manifest. We hypothesized that a multivariate model consisting of graphomotor dysfluency and other pen movement temporal and spatial variables, measured in gene-positive at-risk individuals prior to HD diagnosis, would be associated with decline on a multidimensional measure of HD morbidity.
MATERIALS AND METHODS
Subjects
This study enrolled a subset of subjects from a previously published study of handwriting dysfunction in premanifest HD [13]. Subjects were recruited through the Huntington’s Disease Society of America (HDSA) Center of Excellence at the University of California San Diego. Inclusion criteria for enrollment into this study were as follows: 1) positive test for HTT gene mutation; 2) off medication prescribed for managing a movement disorder; 3) absence of cognitive impairment based on baseline Montreal Cognitive Assessment (MoCA) score less than 24; 4) absence of motor impairment based on total score > 4 on the UHDRS total motor score (TMS) subscale; and 5) an interval between initial and follow-up assessment within 36 months. Seventeen (11 female and 6 male) subjects met criteria for inclusion. Table 1 shows the demographic and clinical characteristics of the 17 study subjects. This research was reviewed and approved by the University of California, San Diego Institutional Review Board (IRB Reference # 170038). All subjects signed informed consent prior to participating in this research.
Demographic and Clinical Characteristics of Study Subjects at Baseline
CAG, repeat in the cytosine-adenine-guanine trinucleotide in the HTT gene; CAP, CAG-age-product score; agex(CAG-30)/6.49; TMS, total motor score of the UHDRS; TUG, timed up and go test; MoCA, Montreal Cognitive Assessment.
Clinical assessments
All subjects underwent clinical assessment of cognitive, motor, and behavioral signs and symptoms at intake. The clinical assessment battery consisted of the Unified Huntington’s Disease Rating Scale (UHDRS) [20], the total motor score from the UHDRS (TMS), the composite UHDRS (cUHDRS) index [21], the timed up and go test (TUG; a measure of gross motor ability and balance) [22], and the MoCA [23]. Confirmatory genetic findings including CAG repeats to assess disease penetrance for PM subjects were available for the study.
The initial assessment following consent and enrollment into the research program was considered baseline for the purpose of this research. The follow-up visit, within the first 36 months following baseline assessment, associated with the greatest decline in cUHDRS was considered the endpoint visit. The average interval between baseline and endpoint was 21.92 (7.40) months.
Graphomotor procedure
Graphomotor assessments were performed at each visit. However, to test the hypothesis that graphomotor performance is an early predictor of motor and functional decline in premanifest HD, only the baseline scores were used. These procedures were described previously [13] and will be briefly summarized here. Graphomotor assessments have been used in several studies from our laboratory to evaluate effects of antipsychotics on motor function [24, 25], as a quantitative instrumental measure of tardive dyskinesia [26] and more recently as predictors of treatment response to long-acting injectable antipsychotics in schizophrenia [27].
The procedure involves natural handwriting and drawing using an electromagnetic stylus (non-inking pen) and a digitizing tablet (WACOM, https://www.wacom.com/en-us). Subjects were instructed to complete the following tasks: simple loops written from left to right (“llllllll”), complex loops (“lleellee”), rapid overlay circles, Archimedes’ spiral, and the sentence “The sky is blue”. All tasks were completed with the subject’s self-reported dominant hand. The tasks were administered in random order with a fixed block of five trials for each. Prior to the start of each task, the subject was shown a visual copy of the task, given instructions, and provided with a practice period to ensure familiarity with writing on the tablet using an inkless stylus. Subjects were instructed to draw as many loops or circles as possible, a spiral and a sentence each within a 10-second recording timeframe and to repeat this over five trials. Trials were separated by a 2-second interval. The graphomotor battery required approximately 10 minutes to complete. MovAlyzeR® software (Neuroscript LLC, Tempe AZ; http://www.neuroscript.net/) was used to record, store, and process each individual pen stroke. Figure 1 shows single trial examples of the graphomotor traces for each of the five tasks.
Single trial examples of the graphomotor traces for each of the five tasks including left-right loops (A), maximum speed circles (B), complex loops (C), spiral (D), and sentence (E). Up and down pen strokes were segmented based on zero velocity axis crossings marked as dots on the raw traces.
Graphomotor data processing
Prior to data processing, graphomotor traces were visually inspected to ensure subjects followed instructions and completed the five trials across five conditions with enough pen strokes. This quality control step resulted in excluding trials and/or conditions for some subjects. Once the graphomotor traces passed quality control, the files were subjected to automated processing to extract graphomotor summary scores from each pen stroke of each condition for each subject.
Automated processing consisted of low pass filtering of the pen movement traces at 12 Hz and calculation of time derivatives (velocity and acceleration). Up and down pen strokes were segmented based on zero velocity axis crossings, reflecting a change in direction (shown in Fig. 1). Kinematic and pressure values were then calculated for each up and downstroke to yield the means and standard deviations averaged across all vertical strokes and trials across five handwriting tasks. Means and standard deviations were calculated from the following variables: 1) movement duration (in milliseconds), 2) vertical amplitude (in centimeters), 3) peak vertical velocity (in centimeters/second), 4) average normalized jerk (a measure of graphomotor smoothness), 5) number of acceleration peaks (a second measure of graphomotor smoothness), and 6) pen pressure (in digital units ranging from 1-1024). Normalized jerk is unitless as it is normalized for stroke duration and length. Average normalized jerk was calculated using the following formula: √(0.5 x ∑(jerk(t)2) x duration5/length2) [28].
Statistical analyses
Univariate procedures
First steps in the statistical analyses involved a series of univariate correlational analyses to examine the simple relationships between graphomotor measures and motor and cognitive change in subjects. Univariate and multivariate (see below) procedures were conducted using Statistica (version 13.3; TIBCO Software, Inc.; http://statistica.io). Pearson correlational tests were performed on each of the 12 graphomotor variables for each of five tasks to evaluate their relationships with baseline TMS, baseline cUHDRS, and baseline MoCA, as well as their respective change scores. Correlation coefficients with p-values≤0.5 were considered statistically significant.
Multivariate procedures
Multivariate linear regression was used to evaluate the utility of our graphomotor measures in predicting multidimensional decline in premanifest HD based on the cUHDRS change score. Change on the cUHDRS was calculated as the difference between baseline and endpoint cUHDRS total score. Separate regression analyses were performed for each of the five handwriting tasks (L-R loops, complex loops, rapidly drawn circles, spirals, and sentences). Independent variables consisted of the means and standard deviations for each of the 6 variables listed above. The 12 independent variables were first entered into a full model and then into a backward stepwise model to evaluate whether the reduced model performed better than the full model. A stopping rule of F < 3.0 was used as a significance threshold for removing variables from the reduced models. Key results from the full and reduced regression models include R2 (the proportion of the variance in the response variable than can be explained by the predictor variables in the model), adjusted R2 (modified R2 adjusted for the number of predictors in the model), standard error of the estimate (the deviation of the predictor estimates from the measured values), F-ratio, and p-value. Models having p≤0.5 are considered statistically significant. The significance level and beta coefficient for graphomotor variables remaining in significant models are also reported.
Statistically significant regression models were subjected to power analyses to estimate the probability of correctly accepting the hypothesis that graphomotor dysfluency and other pen movement temporal and spatial variables, would be associated with decline on a multidimensional measure of HD morbidity. This was deemed particularly important in a small sample study such as this to reduce Type II error. We calculated power for the F-tests associated with the best-fit regression models using G*Power (Version 3.1.9.6; https://www.psychologie.hhu.de/arbeitsgruppen/allgemeine-psychologie-und-arbeitspsychologie/gpower). F-test power calculations were based on the following input parameters: effect size (Cohen’s f =√R2/1-R2); α= 0.05; β=0.95; and number of predictors in the model. Models associated with power > 0.80 were considered protected from false discovery (i.e., Type II error) in this small sample study.
RESULTS
Univariate results
Graphomotor variables were subjected to Pearson correlational analyses to examine their relationships with baseline clinical test scores. Lower pen stroke amplitudes for the L-R loop task were correlated with lower baseline MoCA scores (r = 0.59; p = 0.01). Three variables were associated with lower baseline cUHDRS scores, including reduced stroke peak velocity for spirals (r = 0.65; p = 0.003), increased number of acceleration peaks for spirals (r = –0.63; p = 0.005), and increased stroke duration for spirals (r = –0.73; p < 0.001). None of the graphomotor variables was associated with baseline TMS score after correcting for Type I error. Furthermore, results from simple univariate correlational analyses failed to identify any association between cUHDRS change score and CAG, CAP, age, or any cognitive or graphomotor variable.
Eleven of the 17 subjects exhibited decline (negative change score) on the cUHDRS scale over the study interval with a mean (sd) change of –1.22 (1.57) points.
Multivariate results
Table 2 shows the results from the multiple linear regressions with cUHDRS change score serving as the response variable. Results are presented for each of the five handwriting tasks and include key outcomes from the full models (with 12 variables) and reduced models with number of variables ranging from five to eight). Sample sizes shown in the first column vary due to missing data. The 6th model incorporated the significant variables from two significant models (complex loops and maximum speed circles).
Statistical results from the full and reduced regression models predicting cUHDRS change score. Ful models included 12 graphomotor variables for each task (numbered 1–5); reduced models represented the best fit to measured data based on fewer variables
It is important to note that in all cases, the R2 for the full model is larger than for the reduced model. This is a natural consequence of the statistic and does not imply that the full model is a better fit than its corresponding reduced model. Rather, the adjusted R2 reflects how useful the model is, after adjusting for the number of predictors. Three multivariate regression models based on graphomotor variables derived from a complex loop task, a maximum speed circle drawing task and the combined tasks returned adjusted R2 coefficients of 0.76, 0.71, and 0.80 respectively accounting for a significant portion of the variability in cUHDRS change score. The relatively low standard errors for these models reflect less deviation of the predictor estimates from the measured values compared to other models tested and represents a reasonably good fit of the predictors in accounting for variability in the response variable.
As previously stated, three models associated with decline in cUHDRS score reached statistical significance. Power tests performed on each of the three models revealed probabilities of correctly accepting our hypothesis were 0.67, 0.63, and 0.90 for the model based on complex loops, maximum speed circles, and a combined model, respectively. Thus, only the combined model, based on five variables, yielded an F-statistic with sufficient power to support our hypothesis. To achieve power greater than 0.80, the two underpowered models would require sample sizes of at least 18 subjects.
Table 3 summarizes the regression results for the three significant models. A negative β coefficient indicates that decline on the cUHDRS measure was associated with an increase in the score for that graphomotor variable; whereas a positive β coefficient indicates that decline on the cUHDRS measure was associated with a decrease in the score for that graphomotor variable. Graphomotor factors significantly associated with a decline on the cUHDRS measure consisted of increased pen movement dysfluency (ANJ, acceleration peaks, and pen pressure variability), increased stroke amplitude, decreased stroke-stroke variability on the graphomotor dysfluency, and decreased pen movement velocity. The best-fit model (based on the combined tasks) indicated that greater decline on the cUHDRS was associated with increased pen movement dysfluency and its variability based on the average normalized jerk (ANJ) and the number of acceleration peaks per stroke. A composite graphomotor dysfluency summary score was calculated by multiplying each of the five dysfluency scores with their respective uncorrected β coefficient and summing the products. Using a composite graphomotor dysfluency score change in cUHDRS over a 2-year period (on average) may be estimated using the following equation: cUHDRS change = 5.133 + 0.998 x composite score. Figure 2 shows a scatterplot of the relationship between the baseline composite graphomotor dysfluency score and cUHDRS change score (r = 0.93; p < 0.0001).
Results showing statistically significant graphomotor variables from the three stepwise multiple regression models predicting cUHDRS change score
1SD, standard deviation across strokes and trials; 2 Standardized β coefficient.
Scatterplot of the relationship between the baseline composite graphomotor dysfluency score and cUHDRS change score.
DISCUSSION
The present study examined the association between graphomotor measures of hyperkinesia and hypokinesia and decline on a multidimensional index of HD morbidity in premanifest at-risk individuals. We hypothesized that a multivariate model consisting of graphomotor dysfluency and other pen movement speed and amplitude factors, measured in gene-positive at-risk individuals prior to HD diagnosis, would be associated with decline on the cUHDRS index. Because HD ultimately presents as a hyperkinetic movement disorder, we expected the graphomotor dysfluency variables to be significant factors in our predictive models. Based on analysis of these models and after adjusting for the number of predictors in the models, our findings support this hypothesis in a relatively small sample of highly selected asymptomatic subjects at risk for HD. The high degree of statistical significance for the most robust model underscores the clinical utility of graphomotor measures as a behavioral biomarker of the clinical state in HD.
The most robust predictive model accounted for 80% of the response variability and involved scores from five measures of handwriting dysfluency. Of the five handwriting tasks administered to each subject, two tasks produced graphomotor variables that contributed to this highly significant model: 1) drawing complex loops, with alternating “ll” and ‘ee” and 2) drawing overlay circles as fast as possible. The five dysfluency variables consisted of: ANJ and variability in the number of acceleration peaks (from the first task) and ANJ, number of acceleration peaks, and variability in the number of acceleration peaks (from the second task).
With a power score of 0.90, this model supports the advantage to research of utilizing quantitative instrumental measures of motor dysfunction over traditional observer-based severity rating scales, especially in small-sample studies. Handwriting dysfluency has been observed in individuals with putative hyperdopaminergia such as psychosis [29] and tardive dyskinesia [26]. Similarly, Medzech and colleagues [30] reported abnormalities on their measure of hand force instability, in a group of HD patients. These prior findings and the results of the present study suggest that instrumental measures of hand fine motor control have potential as sensitive behavioral biomarkers of subclinical hyperdopaminergia in premanifest HD.
Many of the relevant large-scale studies on predictors of decline or time to diagnosis in premanifest HD, such as PREDICT-HD focused on cognitive variables and to a lesser extent, neuroimaging findings. Significant predictor from these studies included baseline severity of motor impairment [8, 12], CAG [10, 12], CAP [10], and performance on several cognitive measures [7–9, 11]. Where reported, the R2 ranged from 36% [10] to 51% [7] leaving a large portion of the response variability unaccounted for. Extensive within and between individual variability in the initial presentation of motor impairment has been reported [31, 32], making traditional observer-based assessments of motor function unfeasible for monitoring disease progression in the premanifest stage of HD. Objective quantitative measures such as handwriting kinematics could benefit future research by serving as a baseline screening measure for enrolling more appropriate candidate subjects into clinical trials.
The association between graphomotor dysfluency and disease progression prior to confirmed diagnosis is consistent with survey results from Kirkwood et al. [33] who reported that involuntary movements were the earliest symptom to present within the first year of diagnosis in 36% of study respondents. It is possible that individuals who exhibit subclinical motor impairment (e.g., on measures of graphomotor dysfluency) prior to diagnosis may progress more rapidly to onset of overt involuntary movements than those who do not exhibit subclinical motor impairment. To test this hypothesis, it will be important to follow these individuals through diagnosis and to carefully characterize their progression of symptoms. Confirmation would inform enrollment into clinical trials aimed at delaying the onset of overt motor symptoms in premanifest HD. This information could help in the design of more cost-effective clinical trials.
The current study has limitations. First, as demonstrated by the power analyses, even after reducing the number of independent variables some of our predictive models remained underpowered, despite being statistically significant. The sample size estimates for these models indicated that by increasing the sample size to 18, power would have reached 0.80. Nonetheless, it is important not to overinterpret the results from these underpowered analyses. Rather, these results should be considered exploratory at this time. That multiple regression analyses with up to eight independent variables could achieve sufficient statistical power with only 18 subjects underscores the research value of quantitative motor measures. Quantitative instrumental measures of hand motor impairment have been shown in prior studies to be associated with greater sensitivity to subclinical motor impairment and lower within-subject variability leading to larger effect sizes than conventional observer-based rating scales [34]. Secondly, our follow-up period was also brief relative to other longitudinal studies. As this study was conducted in a clinical environment, scheduling of follow-up assessments was driven by clinical convenience rather than a strict research schedule of events. Large sample replication studies are needed to address these limitations.
In conclusion, to our knowledge, this is the first study to consider graphomotor variables in a longitudinal study design to predict motor or functional decline in premanifest HD. A set of graphomotor kinematic dysfluency variables, purported to be sensitive to subclinical hyperkinesia stemming from dysregulation within the basal ganglia motor circuit, accounted for 80% of the variability in decline on disease morbidity index comprised of motor, cognitive, and functional measures in premanifest HD. Replication of this research is an important step in supporting the utility of instrumental graphomotor assessments to enrich enrollment in clinical trials designed to delay the onset of overt motor symptoms in premanifest HD.
Footnotes
ACKNOWLEDGMENTS
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Authors thank the patients, families, and volunteers from the UC San Diego Huntington’s Disease Society of America (HDSA) Center of Excellence and the UC San Diego Shiley-Marcos ADRC for their participation in this research.
CONFLICT OF INTEREST
The authors have no conflicts of interest to report.
DATA AVAILABILITY
The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy, ethical restrictions, or other concerns.