Tremor in cervical dystonia

Abstract

Background: Cervical dystonia (CD) is the most common form of focal dystonia encountered in the clinic. Approximately one-third of CD patients have co-existing tremor in the head and hands. Assessment of tremor as regular or irregular in context of its oscillation trajectory, frequency, and amplitude is a major clinical challenge and can confound the diagnosis of CD. The misdiagnosis may lead to therapeutic failures, poor quality of life, and poor utilization of medical and financial resources.

Methods: We analyzed the largest cohort of CD patients (n = 3117) available to date, collected from 37 movement disorder centers in North America, Europe, and Asia. We used machine learning to determine what clinical features from clinician reports predicted the presence of tremor as well as its regular or irregular appearance.

Results: Out of 3,117 CD patients, 1,367 had neck tremor. The neck tremor was interpreted as irregular in 1,022, regular in 345, and mixed (both irregular and regular) in 442. A feature importance analysis determined that greater severity of CD, longer disease duration, and older age, in descending order, predicted the presence of neck tremor. The probability of neck tremor was reduced if the dystonia affected other body parts in addition to the neck. We also found a significantly heightened risk for developing neck tremor in women. An additional feature importance analysis indicated that increased severity of dystonia affecting other body parts, severity of CD, and prolonged disease duration was associated with a lower likelihood of regular neck tremor while increased age predicted a higher likelihood.

Conclusion: Machine learning recognized the most relevant clinical features that can predict concurrent neck tremor and its irregularity in a large multi-center dystonia cohort. These results may facilitate a more accurate description of neck tremor and improved care path in CD.

Keywords: cervical dystonia; dystonia; jerkiness; reularity; tremor.

FIGURE 1

Disparity in reported prevalence rate of tremor in patients who have dystonia (A) and prevalence rate of dystonia in those who have tremor (B).

FIGURE 2

A The totals for individual body regions with dystonia (A) and tremor (B) in 3117 participants with cervical dystonia. The totals may sum up to more than the total number of participants because many participants had more than one body region affected. The numbers in the figures show the actual numbers of participants with each region affected.

FIGURE 3

(A) Features relevant for neck tremor in the CD population. (B,C) Features relevant for neck tremor in (B) female and (C) male patients. Significant features predicting tremor were determined by Wald tests. Significant parameters (shown in red) are significantly different from zero and produce a statistically significant decline in the logistic regression model once removed. The impact of each parameter is estimated by the length of the line. Non-significant features are shown in black.

FIGURE 4

Clustering analyses results. (A–C) Clustering of female patients. (A) The Elbow method was used to find the optimum number of clusters, k. Within-cluster sum of squares (the sum of squared distance between each point and the centroid in a cluster) are plotted for a range of number of clusters (k = [1,8]). At k = 5, the slope of the graph changes, creating an elbow shape. This point was considered to be the optimal number of clusters for the female group. (B) The three-dimensional scatter plot displays the first three principal components of the five clusters detected by the k-means clustering algorithm (for interactive plot: https://chart-studio.plotly.com/~sinoscope/125). Cluster sizes and neck tremor prevalence rates of the clusters are shown in the legend. (C) Boxplots from left to right show age, dystonia duration, CD severity (GDRS neck) and non-CD severity (GDRS other) distributions of the 5 clusters. An asterisk above a box indicates a statistically significant pair-wise difference between that box and all the others except for the pairs marked with “ns.” (D–F) Clustering of male patients. (D) k = 5, where the graph makes an elbow shape, was considered to be the optimal number of clusters for the male cohort. (E) The three-dimensional scatter plot displays the first three principal components of the four clusters detected by the k-means clustering algorithm (for interactive plot: https://chart-studio.plotly.com/~sinoscope/131). Neck tremor prevalence rates of the clusters are shown in the legend. (F) Boxplots (from left to right) demonstrate age, dystonia duration, CD severity (GDRS neck) and non-CD severity (GDRS other) distributions of the 5 clusters. An asterisk above a box indicates a statistically significant pair-wise difference between that box and all the others except for the pairs marked with “ns.”

FIGURE 5

Features significantly distinguish (A) regular from irregular neck tremor, and (B) regular neck tremor from no tremor. Significant features (shown in red) are significantly different from zero as tested using Wald tests and produce a statistically significant decline in the logistic regression model once removed. The impact of each parameter is estimated by the length of the line. Non-significant features are shown in black.