Temporal Dynamics of Parkinson's Disease Tremor: Clinical and Neuroimaging Insights

Abstract

Background

Parkinson's disease (PD) is a progressive neurodegenerative disorder clinically defined by three cardinal motor symptoms: bradykinesia, rigidity, and tremor. Although the natural history of bradykinesia and rigidity is well described, the evolution of tremor as the disease progresses remains controversial.

Objectives

The goal was to clinically characterize long‐term trajectories of rest, postural, and action tremor in PD and examine their neural network correlates using longitudinal resting‐state functional magnetic resonance imaging (rs‐fMRI).

Methods

In this retrospective longitudinal cohort study, we analyzed 93 tremor‐positive PD patients (mean disease duration: 3.6 years), each with up to six clinical assessments over 4.2 years. Linear mixed‐effects models assessed temporal change in Movement Disorder Society‐Unified Parkinson's Disease Rating Scale (MDS‐UPDRS) and Fahn–Tolosa–Marín Tremor Rating Scale (FTM‐TRS) scores. Rs‐fMRI data from 30 tremor‐affected patients (mean interval, 3.35 years) were analyzed for seed‐to–whole‐brain connectivity and connectivity between predefined brain regions.

Results

Bradykinesia (β = +0.09, P < 0.001) and rigidity (β = +0.06, P < 0.001) worsened, whereas total FTM‐TRS scores declined by 0.49 points/year (P = 0.037), driven by reductions in postural (P < 0.001) and action tremor (P = 0.033); resting tremor remained stable. Rs‐fMRI revealed longitudinal changes in cerebellar‐ and thalamic‐seed‐to‐whole‐brain connectivity.

Conclusions

Tremor in PD evolves along distinct clinical courses, often stabilizing or improving as other motor features worsen. These findings are consistent with partially adaptive reorganization within tremor‐related networks and underscore the importance of identifying patient subtypes with divergent trajectories to inform prognosis and optimize therapy. © 2026 The Author(s). Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

Parkinson's disease (PD) is one of the most prevalent neurodegenerative disorders globally, characterized by motor symptoms such as bradykinesia, rigidity, and tremor. ¹ , ² Among these, tremor stands out for its variability, inconsistent treatment response, unclear progression and mechanisms. ³ , ⁴ , ⁵ , ⁶

Longitudinal studies on tremor progression in PD have produced mixed results. Over 7 years, one study found significant increases in off‐state rest, postural, and kinetic tremor scores, ⁵ whereas another ⁶ observed stable resting tremor, reductions in postural and kinetic tremor and worsening bradykinesia and rigidity. Similarly, a recent study ⁷ reported decreasing tremor severity, most prominently in action tremor.

These conflicting findings highlight the need for longitudinal studies that more accurately capture the evolution of tremor and determine whether distinct patterns of progression exist.

Most longitudinal studies have relied on the Movement Disorder Society‐Unified Parkinson's Disease Rating Scale (MDS‐UPDRS), ⁸ which is limited in differentiating tremor types, as it includes only one item each for resting (3.17), postural (3.15), and action tremor (3.16). To address this, we additionally applied the Fahn–Tolosa–Marín Tremor Rating Scale (FTM‐TRS). ⁹ The FTM‐TRS captures a broader range of tremor manifestations across body regions and tasks such as handwriting and pouring. Its three parts evaluate tremor severity (A), motor performance (B), and tremor impact on activities of daily living (C), providing a more comprehensive and valid assessment, particularly for action tremor.

The FTM‐TRS is a sensitive and widely accepted tool for assessing tremor severity and longitudinal change. The MDS Task Force on Tremor Rating Scales recommended it as a reliable and “sensitive to change” measure, emphasizing its superior clinometric properties. ¹⁰ In a head‐to‐head comparison, one study showed that the FTM‐TRS was more responsive to treatment effects than MDS‐UPDRS tremor items. ¹¹ By combining both scales, our study captures a more detailed and reliable trajectory of rest and action tremor progression in PD.

Resting‐state functional magnetic resonance imaging (rs‐fMRI), a magnetic resonance imaging (MRI) method that measures spontaneous brain activity during rest to reveal functional connectivity (FC) between brain regions, has greatly advanced understanding of network alterations in PD by mapping FC changes. Converging evidence links PD tremor to interactions between basal ganglia and the cerebello‐thalamo‐cortical (CTC) motor loop, with FC alterations in this circuit correlating with tremor severity and treatment‐related suppression. ¹² A recent study ¹³ further demonstrated retest stability of clinical, neurophysiological, and rs‐fMRI tremor measures with consistent CTC engagement, supporting rs‐fMRI as a reliable systems‐level biomarker. However, longitudinal rs‐fMRI studies, particularly those examining tremor trajectories, remain scarce, highlighting the need for extended follow‐up and harmonized designs. ¹⁴

The present study aims to integrate longitudinal MDS‐UPDRS and the more granular FTM‐TRS assessments to enable finer stratification of tremor subtypes, particularly action tremor. In a subset of patients, longitudinal rs‐fMRI was incorporated to relate clinical trajectories to CTC network dynamics. This dual approach addresses critical gaps in tremor phenotyping and imaging, providing novel insights into the progression and underlying mechanisms of PD tremor.

Patients and Methods

Participants

Data from 118 PD patients was obtained from the Prospective Movement Disorders Registry Graz (PROMOVE) database, a single‐center cohort at the Department of Neurology, Medical University of Graz, Austria. Established in 2010, PROMOVE provides standardized, prospectively collected clinical and imaging data for longitudinal research in movement disorders, with follow‐up visits every 12 to 18 months for up to six time points.

Inclusion required a clinical PD diagnosis according to the United Kingdom Parkinson's Disease Society Brain Bank criteria. ¹⁵ All participants also fulfilled the MDS Task Force diagnostic criteria, ¹⁶ applied consistently since 2015. Dopamine transporter imaging (DAT‐SPECT) was performed at baseline in 55 patients (48.7%) and showed abnormal results in all of them. Exclusion criteria were significant cognitive impairment (Mini‐Mental State Examination [MMSE] <24), major psychiatric illness, or other neurological disorders affecting motor function. All participants provided written informed consent, and the study was approved by the ethics committee of the Medical University of Graz (21‐345 ex 09/10).

At each visit, participants underwent a comprehensive assessment covering motor and non‐motor domains as well as cognitive scores, including the FTM‐TRS, MDS‐UPDRS, Non‐Motor Symptoms Questionnaire (NMS‐Quest), ¹⁷ Hoehn and Yahr staging (H&Y), ¹⁸ Consortium to Establish a Registry for Alzheimer's Disease battery (CERAD‐Plus battery), ¹⁹ Geriatric Depression Scale (GDS), ²⁰ MMSE, ²¹ and the Frontal Assessment Battery (FAB). ²²

Examinations were performed in the “practically defined off” state after at least 12 hours of dopaminergic medication withdrawal.

MDS‐UPDRS part III scores quantified bradykinesia (items 3.4–3.11, 3.14), rigidity (item 3.3), and tremor (items 3.15–3.17). Tremor subtypes were defined by amplitude subscores for resting (3.17), postural (3.15), and action tremor (3.16), normalized to a 0 to 4 scale. For FTM‐TRS part A, mean scores (0–4) were computed for rest (8 items), postural (7 items), and kinetic tremor (5 items). Tremor laterality was defined at baseline using unilateral upper extremity FTM‐TRS/MDS‐UPDRS tremor subscores. For each patient, left‐ and right‐sided tremor burden was calculated as the sum of resting, postural, and action tremor scores. ¹³ , ²³ The most‐affected side (MAS) (side with most tremor points overall) and least‐affected side (LAS) (side with least tremor points overall) were defined accordingly. Baseline laterality classification was used for all subsequent analyses.

At baseline (n = 112), bradykinesia was present in all patients, rigidity in 108, and tremor in 96 (rest = 76, kinetic = 73, postural = 71). Longitudinal analyses included only patients with an MDS‐UPDRS tremor subscore >0 at baseline. Patients without tremor were excluded from clinical and fMRI analyses but summarized descriptively in the Data S1. Tremor severity was further evaluated using FTM‐TRS parts A–C (amplitude, task performance, and disability; 0–4 scale), analyzed separately.

Participant inclusion, exclusion, and attrition are shown in Figure 1. Participants completed on average 3.63 follow‐ups (range = 1–5), yielding 418 inter‐visit intervals. Visit spacing averaged 12.9, 14.7, 13.8, 18.2, and 14.4 months between consecutive visits, respectively. The mean inter‐assessment interval across all visits was 14.7 ± 7.0 months, and the mean total observation period was 50.7 months (4.2 years; range = 19–73, standard deviation [SD] = 15).

FIG. 1.

Patient selection process. Patient selection flow‐chart showing sequential exclusions from the initial Prospective Movement Disorders Registry Graz (PROMOVE) cohort (n = 118) to the final tremor dataset (n = 93) after removing cases for diagnosis change, missing data, absent baseline tremor, and one outlier. [Color figure can be viewed at wileyonlinelibrary.com]

MRI

High‐resolution brain MRI, including T1‐weighted and rs‐fMRI, was acquired using a standardized protocol. Because of a scanner replacement in 2015, causing cross‐session inconsistencies, only post‐replacement scans were analyzed. To ensure comparability, rs‐fMRI analyses were limited to two time points per participant, the earliest and latest post‐replacement scans. Participants with only one eligible scan were excluded, yielding 44 patients with two rs‐fMRI sessions on the same scanner. Given variable interscan intervals, MRI and behavioral measures were annualized as described in the Statistical Analysis section. MRI was performed on a 3 Tesla Prisma scanner (MAGNETOM Prisma fit, Siemens Healthcare, Erlangen, Germany) using a 20‐channel head coil. Detailed information on MRI can be found in Data S1.

Functional Imaging Analysis

The T1‐weighted and resting state images of 44 patients were preprocessed using harmonized analysis of functional MRI pipeline (HALFpipe) version 1.2.3., ²⁴ which is based on fMRIPrep, ²⁵ following standardized protocols to harmonize analysis and quality control (scans with mean framewise displacement >0.3 mm, >20% high‐motion volumes [Framewise Displacement (FD) > 0.5 mm], or temporal signal‐to‐noise ratio [SNR] < 40 were excluded) across multiple sites (see https://enigma.ini.usc.edu/protocols/functional‐protocols/). Preprocessing included motion correction, slice‐timing correction, co‐registration, spatial normalization to Montreal Neurological Institute (MNI) 152 NLIN 2009c space, and 6‐mm smoothing. Denoising combined aCompCor with ICA‐AROMA to remove motion and physiological artifacts. A detailed description on preprocessing and settings can be found in Data S1.

Twelve subjects were excluded for absent baseline tremor (n = 5) or poor data quality (n = 7) and two for a revised diagnosis, leaving 30 patients for analysis. Imaging data was used from this subsample, two high‐quality scans from each time point. The mean inter‐scan interval was 3.35 years (SD = 1.56; range = 0.92–6.41). Rs‐fMRI analyses targeted the CTC, following Angelini et al. ¹³ As can be seen on Figure 2, regions of interest (ROIs) were defined a priori, including the ventrolateral posterior thalamus (Fig. 2A, Nucleus ventralis posterolateralis [VLp]; Morel Atlas), the upper‐limb representation of primary motor cortex (Fig. 2B, Area 4 upper limb (A4ul); Brainnetome Atlas) and cerebellar lobules V–VI (Fig. 2C, Oxford Centre for Functional MRI of the Brain Software Library [FSL]; Cerebellar Atlas). Masks were generated in FSL and resampled to the HALFpipe template. Detailed description on mask generation and analysis can be found in Data S1.

FIG. 2.

Top (A–C): regions of interest (ROIs) within the cerebello‐thalamo‐cortical circuit (CTC). Magnetic resonance imaging (MRI) overlays identifying thalamic ventrolateral posterior nuclei (A) Morel Atlas, primary motor cortex upper‐limb representations (B) Brainnetome Atlas and cerebellar lobules V/VI (C) FSL Atlas. ROIs are shown bilaterally across sagittal, coronal, and axial views. Bottom (D–F): Longitudinal changes in seed‐based functional connectivity. Significant clusters were identified using threshold‐free cluster enhancement (TFCE)‐based permutation testing (P < 0.05, family‐wise error [FWE]‐corrected). For the cerebellar seed, decreased functional connectivity (FC) was detected in the putamen (D), whereas increased FC was observed in the lingual and occipital fusiform gyrus (E). For the thalamic seed, a significant increase in FC was found in the precuneus cortex and postcentral gyrus. No significant longitudinal effects were observed for the motor seed. [Color figure can be viewed at wileyonlinelibrary.com]

To account for differences in symptom lateralization, all preprocessed fMRI datasets of participants with a right‐sided MAS (n = 11) were flipped. ¹³ This ensured a consistent hemispheric alignment across subjects, with the left hemisphere representing the ipsilateral and the right hemisphere the contralateral side relative to the MAS. All subsequent connectivity analyses were performed in this standardized orientation.

Statistical Analysis

Analyses were performed in R (version 4.5.1) using linear mixed‐effects models (package lme4) to assess longitudinal changes in clinical and cognitive outcomes within the tremor group (n = 93). Models were fitted with restricted maximum likelihood and included time (years since baseline), sex, and baseline disease duration (DD) as fixed effects, and subject‐specific intercepts and slopes for time as random effects. Time was modeled continuously based on actual assessment dates to account for unequal follow‐up intervals. In cases of singular fits, the random‐effects structure was simplified to random intercepts only. This was the case for bradykinesia (MDS‐UPDRS), FTM drinking and pouring (FTM), and kinetic tremor (MDS‐UPDRS). In a sensitivity analysis, this was replicated on the whole patient dataset without exclusion of patients with no tremor at baseline. Between‐group differences (tremor vs. no‐tremor) were evaluated with unpaired t tests. More information on analysis and the full model code is available in Data S1.

Imaging Data

Seed‐Based Analysis

For each previously described seed, whole‐brain functional connectivity maps were computed using FSL's dual_regression approach. Group‐level statistical inference was performed using non‐parametric permutation testing with 5.000 permutations. Multiple‐comparison correction was applied using threshold‐free cluster enhancement (TFCE).

ROI × ROI Analysis

ROI‐to‐ROI functional connectivity was computed by extracting mean ROI time series from the denoised residual blood oxygen level dependent (BOLD) time series generated by the ENIGMA preprocessing pipeline. These residual time series represent the BOLD signal after nuisance regression, including motion‐related regressors and physiological noise components. Pairwise Pearson correlations were then calculated between all ROI pairs, resulting in one correlation matrix per participant and session. Correlation coefficients (r) were Fisher z‐transformed for statistical analysis. For each ROI pair, the mean difference (T2–T1) represents the change in Fisher z‐transformed connectivity between sessions. Repeated‐measures analysis of covariance (ANCOVA) with time between scans as a covariate for predefined ROIs as well as false discovery rate (FDR)‐correction (Benjamini‐Hochberg) for multiple comparisons (n = 3) was used.

Results

Clinical Data

The tremor group (n = 93; 73 men, 20 women) and the no‐tremor group (n = 16; 12 men, 4 women) did not differ significantly in age, body mass index (BMI), DD, NMS‐QUEST, MMSE, FABT, H&Y stage, or MDS‐UPDRS subscores for rigidity and bradykinesia. However, the no‐tremor group showed higher GDS scores in points (mean [M] = 3.29, SD = 2.29) than the tremor group (M = 2.18, SD = 2.05; P = 0.044) and performed worse on the CERAD memory subscore (z‐score; M = −0.61, SD = 0.83 vs. M = −0.05, SD = 0.73; P = 0.007). A detailed table comparing the two groups in baseline measures is presented in Data S1.

At baseline, tremor severity measured by the MDS‐UPDRS was significantly higher on the MAS (2.89 ± 1.58) compared with the LAS (1.20 ± 1.26; mean difference = 1.69, t(92) = 11.64, P < 0.001), confirming clinical asymmetry. Among the 93 patients with tremor at baseline, 35.5% showed right‐dominant tremor, 40.9% left‐dominant tremor, and 22.6% symmetric involvement, 1 person did not have sufficient data at baseline.

At baseline and across follow‐up visits, lateralization patterns remained broadly stable for both the FTM and MDS‐UPDRS, with approximately one‐third of patients showing right‐dominant, left‐dominant, or symmetric involvement at each time point. More detailed information regarding tremor asymmetry can be found in Data S1.

FTM‐TRS and MDS‐UPDRS

Change in FTM‐TRS and MDS‐UPDRS scores was evaluated using linear mixed‐effects models including all available visits (n = 93). As shown in Table 1, total FTM‐TRS scores decreased significantly over time (β = −0.49 points/year, P = 0.037), indicating an overall reduction in tremor severity. The tremor examination subscale (A) declined by 0.27 points/year (P = 0.007), while in the action subscale (B), handwriting and drinking/pouring scores decreased by 0.04 and 0.18 points/year (P = 0.003 and 0.008, respectively). Functional disability (subscore C) showed no significant change (β = −0.06 points/year, P = 0.360). Both postural tremor (β = −0.031 points/year, P < 0.001 ) and action tremor (β = −0.014, P = 0.033) decreased significantly over time, while resting tremor remained stable (β = −0.006, P = 0.310). Analysis of MDS‐UPDRS subscores showed worsening of rigidity and bradykinesia, increasing by 0.06 and 0.09 points/year (both P < 0.001), whereas the tremor composite showed no statistically significant change (β = −0.02 points/year, P = 0.111) and no significant subtype‐specific effects.

TABLE 1.

Annualized change in tremor and motor severity.

	Slope (β) per yr	P (test β ≠ 0)
FTM‐TRS scores (higher = more severe), tremor group (n = 93)
FTM‐TRS	−0.49	0.037
FTM‐TRS A tremor examination	−0.27	0.007
FTM‐TRS B handwriting	−0.04	0.003
FTM‐TRS B drinking/pouring	−0.18	0.008
FTM‐TRS C functional disability	−0.06	0.360
FTM‐TRS resting tremor	−0.006	0.310
FTM‐TRS postural tremor	−0.031	<0.001
FTM‐TRS action/intention tremor	−0.014	0.033
MDS‐UPDRS scores (higher = more severe), tremor group (n = 96)
MDS tremor	−0.02	0.111
MDS rigidity	0.06	<0.001
MDS bradykinesia	0.09	<0.001
MDS tremor resting	−0.01	0.305
MDS tremor postural	−0.03	0.060
MDS tremor kinetic	−0.03	0.073

Note: Linear mixed‐effects model estimates showing yearly slopes (β) for FTM‐TRS subscores and MDS‐UPDRS motor subscores. Several FTM‐TRS components showed significant longitudinal decreases, whereas rigidity and bradykinesia (MDS‐UPDRS) demonstrated significant worsening over time.

Abbreviations: FTM‐TRS, Fahn–Tolosa–Marín Tremor Rating Scale; MDS‐UPDRS, Movement Disorder Society‐Unified Parkinson's Disease Rating Scale.

Individual trajectories and model‐predicted means for FTM‐TRS total and subscores as well as for the MDS‐UPDRS subscores are displayed in Figure 3. Gray lines represent predicted scores for each participant (baseline = 1 to FU5 = 6; n = 93), and red lines depict the mean mixed‐model trajectory. Note that the relatively uniform slopes in some plots result from random‐intercept‐only models, which constrain individual rates to the group mean.

FIG. 3.

Longitudinal clinical motor trajectories. Individual participant trajectories (grey) and group mean ± confidence band (orange) for FTM‐TRS total and subscores, and for MDS‐UPDRS subscores. Panels illustrate variability and overall trends across tremor examination, handwriting, functional tasks, resting, postural, and kinetic tremor, as well as rigidity and bradykinesia progression. Abbreviations: FTM‐TRS, Fahn–Tolosa–Marín Tremor Rating Scale; MDS‐UPDRS, Movement Disorder Society‐Unified Parkinson's Disease Rating Scale. [Color figure can be viewed at wileyonlinelibrary.com]

fMRI Results

Seed to Whole Brain Analysis

As can be seen in Figure 2 (bottom) and Table 2, regarding the cerebellar seed, decreased connectivity over time could be observed with the right putamen (Fig. 2D) (cluster‐size [k] = 23 voxel), whereas a larger cluster encompassing the lingual and occipital fusiform gyri exhibited increased connectivity at follow‐up (Fig. 2E) (k = 565 voxel). For the thalamic seed, an increase in connectivity was found in a cluster spanning the precuneus and postcentral gyrus (Fig. 2F) (k = 32 voxel). No significant longitudinal effects were detected for the seed of the primary motor area. No correlations between seed‐based connectivity changes and changes in clinical scores remained significant after FDR‐correction.

TABLE 2.

Seed‐to‐whole‐brain connectivity.

Seed	Direction of change	Brain regions	k (voxels)	x	y	z
Cerebellum	Decrease	Putamen	23	63	66	35
Cerebellum	Increase	Lingual gyrus, occipital fusiform gyrus	565	59	33	37
Thalamus	Increase	Precuneus cortex, postcentral gyrus	32	48	47	53

Note: Significant longitudinal changes in seed‐based whole‐brain functional connectivity identified using TFCE‐corrected permutation testing. All results are TFCE FWE P < 0.05, k denotes cluster size in voxels, and coordinates are given in MNI space.

Abbreviations: TFCE, threshold‐free cluster enhancement; FWE, family‐wise error; MNI, Montreal Neurological Institute.

Between ROI Region Connectivity

As can be seen in Table 3, cerebellar‐thalamic connectivity showed a significant decrease from pre (M = 0.55, SD = 0.05) to post (M = 0.43, SD = 0.04, P = 0.045, ηp² = 0.136). No significant longitudinal effects were found for cerebellar‐motor or motor‐thalamic connectivity. The decrease in cerebellar‐thalamic connectivity was significant at the uncorrected level (P = 0.045), but did not survive FDR correction across the three tests (q = 0.135). No correlations between clinical variables and imaging‐derived connectivity measures remained significant after FDR correction.

TABLE 3.

ROI‐to‐ROI connectivity.

Connection	TP1 mean ± SD	TP2 mean ± SD	Δ (TP2–TP1)	P (FDR)	ηp2, effect size
Cerebellum × thalamus	0.55 ± 0.05	0.43 ± 0.04	−0.12	0.045 (0.135)	0.136
Cerebellum × motor	0.51 ± 0.05	0.47 ± 0.04	−0.04	0.461 (0.692)	0.020
Motor × thalamus	0.31 ± 0.06	0.30 ± 0.05	−0.01	0.792 (0.792)	0.003

Note: Longitudinal changes in pairwise functional connectivity within the cortico‐thalamo‐cerebellar motor circuit. Values represent mean Fisher z‐transformed correlation coefficients ± SD. Δ indicates the mean change from TP2–TP1. FDR‐corrected and raw P‐values as well as partial ηp² are reported.

Abbreviations: SD, standard deviation; FDR, false discovery rate; TP, time point.

Discussion

In this longitudinal cohort study, we observed a modest, but statistically significant clinical reduction in postural and kinetic tremor severity over time, primarily detected by the FTM‐TRS, but also reflected as a statistical trend in the MDS‐UPDRS. Resting tremor remained stable over an average follow‐up period of 4.22 years in both scales. The progression of bradykinesia and rigidity seemed to worsen consistently, as has been shown by several other studies. ⁵ , ⁶ , ²⁶

Our findings align with previous intermediate‐duration studies that reported stable or even declining tremor severity. ⁶ , ²⁷ However, they diverge from longer‐term analyses such as the 7‐year follow‐up done on the Parkinson's Progression Markers Initiative (PPMI) cohort, ⁵ which indicated a progressive increase in all three tremor types.

Notably, prior longitudinal studies predominantly relied on MDS‐UPDRS alone. Our combined longitudinal FTM‐TRS and imaging framework, therefore, offers a finer clinical trajectory and a concurrent network‐level readout.

Sample Composition

A key distinction between previous studies concerns whether PD patients without tremor at baseline were included. In our analysis, as in that of the Personalized Parkinson Project (PPP) study, ⁶ only participants with an MDS‐UPDRS tremor score >0 were modeled longitudinally, isolating true symptom progression from new‐onset tremor. In contrast, the PPMI study ⁵ included all patients regardless of baseline tremor status and examined individuals at very early disease stages, when tremor may not yet have manifested. Converters, patients developing tremor during follow‐up, might have inflated mean scores and created the impression of uniform progression, especially in models without random slopes, even if tremor in already affected individuals remained stable or improved. Both studies used off‐state patient data for their tremor trajectory analysis.

DD and Heterogeneity

Moreover, DD could play a significant role in the different results. Although our cohort had a mean DD at baseline of 3.6 years (SD = 3.62), the PPP‐study had a mean DD of 2.7, ⁶ and another recent study with similar results that of 5.7 years. ⁷ The PPMI‐study had one of 6.7 months. ⁵ This suggests that in very early stages, tremor might not have developed yet.

What makes this more complicated is that the subsection of our cohort that did not show any tremor at baseline (n = 17) showed a DD of 5.3 years (SD = 5.07), which seems to be longer rather than shorter and might indicate a PD subtype. Moreover, based on our data, we would expect tremor scores from the PPMI‐study ⁵ to drop at approximately year 4 or 5. However, it steadily increased until and including year 7.

Although DD at baseline was included as a covariate in our multilevel models, it did not exert a significant effect on any of the outcome measures. In our dataset, analyses including patients without baseline tremor showed similar directional trends to the tremor‐group analysis, but did not reach statistical significance (see Supporting Information Results). These observations emphasize the need to distinguish true progression from new‐onset tremor and highlight the clinical heterogeneity of tremor in PD. Our findings align with those of the PPP‐study, ⁶ which, in their whole cohort analysis (n = 438; 191 without resting tremor at baseline), observed an overall increase in resting tremor over time, primarily driven by 59 patients who developed tremor at 2‐year follow‐up. Their results for the decrease of kinetic tremor remained unchanged when including all subjects.

The inter‐individual variability in patients trajectories likely reflects both clinical and biological diversity within the PD tremor phenotype. Such heterogeneity may also explain findings by a recent study done on long‐term bilateral subthalamic nucleus deep brain stimulation (STN‐DBS) patients ²⁸ who reported progressive worsening of action tremor in their cohort. It is conceivable that individuals with more pronounced or progressive action tremor are preferentially selected for STN‐DBS, representing a distinct clinical subtype within PD.

Laterality represents an additional consideration. One study noted that contralateral spread is common when tremor is initially asymmetric and clinically significant. In de novo cohorts, contralateral tremor emerges in approximately one‐quarter of patients by year 3 and over one‐third by year 7. ⁵ Importantly, tremor on the initially more affected side does not necessarily worsen further, while the contralateral side “catches up,” often alongside bilateral progression of bradykinesia and rigidity. This pattern may weaken correlations between global progression metrics and patients' perceived tremor worsening, underscoring the need for side‐ and tremor‐type‐specific longitudinal endpoints.

FTM‐TRS and Subscores

Although FTM‐TRS parts A (tremor severity) and B (performance/tasks) improved in our cohort, part C (activities of daily living [ADL]) showed no significant change. Similarly, one study ¹¹ found the largest effect sizes in the objective components (parts A/B), and the MDS Task Force ¹⁰ noted that ADL scores such as FTM‐TRS‐C require further validation and are generally less sensitive to change. This likely reflects that part C captures multifactorial disability rather than tremor alone.ADLs in PD are shaped by bradykinesia, rigidity, gait, non‐motor symptoms, and compensatory strategies, so tremor improvements may not translate linearly into functional gains. Moreover, patients may find it difficult to clearly distinguish between disability caused by tremor and that caused by bradykinesia or rigidity, which could further obscure ADL‐specific changes over time.

In our mixed‐effects model, patients showed a mean decline of 0.49 FTM‐TRS points per year, equivalent to a 2.07‐point (≈19.7%) reduction over 4.2 years, relative to baseline values (M = 10.52, SD = 9.88). Given the fact that a 20% change corresponds to a moderate improvement in tremor severity, this can therefore be regarded as clinically meaningful.

fMRI

In the rs‐fMRI subsample, seed‐to–whole‐brain analyses revealed longitudinal connectivity changes for cerebellar and thalamic seeds, whereas the primary motor area seed showed no detectable effects. The cerebellar seed showed reduced coupling with the putamen, consistent with cerebello‐striatal interactions that contribute to parkinsonian motor phenotypes and vary with disease state, ²⁹ alongside increased connectivity with lingual/occipital fusiform regions, paralleling reports of reduced subcortical coupling and increased visual‐association connectivity of the sensorimotor cerebellum in PD. ³⁰ The thalamic seed showed increased connectivity with precuneus/postcentral cortex, in line with heightened coupling between motor thalamic subdivisions and cortical targets in PD. ³¹

Mechanistically, CTC circuitry is proposed to modulate tremor amplitude (the “dimmer”), whereas basal ganglia‐thalamo‐cortical loops may contribute to triggering/terminating tremor episodes (the “switch”). ³² , ³³ Because action/postural tremor depends on continuous sensorimotor integration and cerebellar output to thalamus during voluntary control, CTC alterations may be particularly relevant to action tremor, consistent with evidence for CTC involvement in parkinsonian tremor and dopamine‐resistant phenotypes. ⁴

Consistent with this model, the observed longitudinal reconfiguration of cerebellar and thalamic connectivity may reflect aspects of tremor‐related network adaptation, but given the lack of significant correlations after correction, these imaging results should be viewed as hypothesis‐generating rather than confirmatory. Therefore, clinical and imaging trajectories should be viewed as potentially related, but not directly coupled.

Cerebellar FC changes may also partly reflect progressive atrophy, as longitudinal cerebellar gray‐matter loss has been linked to changes in action tremor severity in PD, ⁶ suggesting some FC alterations could be secondary to tissue degeneration. Although the cerebello‐thalamic decrease in FC was FDR‐nonsignificant, its direction warrants testing in larger samples. Larger harmonized multicenter cohorts with repeated, phenotype‐resolved tremor assessments will be needed to stratify rest‐dominant versus action‐dominant tremor and map subtype‐specific circuit trajectories onto clinical change.

Medication and Practical Off State

Moreover, the observed reduction in action and postural tremor could reflect long‐term medication effects. ³⁴ Overnight withdrawal may be insufficient to eliminate residual dopaminergic influences, ³⁵ and these effects may persist longer for tremor than for bradykinesia/rigidity. However, a more likely explanation is that the progression of bradykinesia and rigidity, as well as the decrease of action and postural tremor, are a result of disease progression and neurodegeneration. We observed decreasing cerebello‐thalamic connectivity and reduced region‐based mean connectivity within cerebellar regions, paralleling less clinical tremor.

To combat this, it could be useful to integrate continuous digital tremor sensing to better disentangle disease‐related network changes from medication effects. The WATCH‐PD study exemplifies this potential, showing that wearable sensors can quantify tremor and related motor features with high sensitivity for tracking progression inside and outside the clinic. ³⁶

Limitations

Although all participants met the MDS Task Force diagnostic criteria for PD throughout the follow‐up period, no genetic testing or additional biomarker‐based subgrouping was done. Another limitation is the incomplete follow‐up data. Not all patients attended all six assessments, and the time between visits varied considerably. Participants with between one and five follow‐ups were included, resulting in unequal follow‐up lengths. Regarding tremor assessment, we were unable to distinguish re‐emergent tremor from pure resting tremor, because the MDS‐UPDRS and FTM‐TRS do not capture this feature separately. Moreover, for some outcomes (bradykinesia, FTM drinking and pouring, and MDS kinetic tremor), singular fits necessitated random‐intercept–only models, which constrains estimation of inter‐individual differences in longitudinal change. Additionally, it should be noted that the null findings for DD should be interpreted cautiously, as DD is inherently correlated with time since baseline, introducing potential collinearity, and its restricted range in our cohort may have further limited sensitivity to detect independent association. Given the absence of an effect of DD and the strong coupling of levodopa equivalent daily dose (LEDD) with disease progression and time, we did not include LEDD as a primary covariate to avoid additional collinearity and loss of statistical power.

In terms of imaging, the rs‐fMRI subsample was modest (n = 30), which precluded analyses stratified by tremor phenotype (rest‐ vs. action/postural‐dominant) or linking subtype‐specific connectivity changes to clinical trajectories. Follow‐up intervals in the cohort varied (0.92–6.41 years), which may introduce additional variance and complicate interpretation of longitudinal effects. Although inter‐scan interval was included as a covariate in the ROI×ROI analysis, it could not be explicitly modeled in the paired TFCE seed‐to‐whole‐brain framework and, therefore, remains a limitation.

Changes in connectivity showed no significant relationship with alterations in tremor severity, indicating that network reorganization and clinical manifestation may progress independently. This null finding may reflect limited power, temporal dissociations between functional reconfiguration and symptom manifestation, and the distributed, multi‐network architecture underlying tremor, which may be only partly captured by resting‐state metrics. A power analysis, available in Data S1, indicated that detecting longitudinal connectivity effects would require substantially larger samples than available here.

Conclusion

This study introduces a dual longitudinal framework combining FTM‐TRS‐enhanced tremor phenotyping with rs‐fMRI to delineate tremor trajectories and their CTC correlates in PD. Our findings substantiate the evidence that PD tremor follows a course that is distinct from other motor symptoms of the disease as well as different tremor types (e.g. rest, postural, and kinetic), with methodological differences such as inclusion criteria and DD likely contributing to variability across studies.

A key innovation of our approach is the integration of (1) the MDS‐UPDRS with the more granular FTM‐TRS to capture action tremor across multiple body regions and tasks, and (2) longitudinal rs‐fMRI to track CTC network dynamics alongside clinical change. We speculate that altered cerebellar/thalamic connectivity over time attenuates CTC‐mediated tremor amplitude, paralleling clinical improvement in action and postural tremor, but not bradykinesia, rigidity, or resting tremor. Conceptually, this mirrors the mechanism of effective deep brain stimulation, which disrupts CTC oscillations through thalamic inhibition.

By combining sensitive clinical scales with repeated network imaging, we offer a higher‐resolution view of tremor progression that may reconcile inconsistencies in prior work and highlight circuit‐specific therapeutic targets. Future research should adopt harmonized longitudinal designs, stratify by clinical and genetic subtypes, and integrate continuous digital tremor sensing to better disentangle disease‐related network changes from medication effects.

Author Roles

(1) Research Project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript: A. Writing of the First Draft, B. Review and Critique.

Melanie. S.: 1C, 2A, 2B, 2C, 3A.

R.M.: 2A, 2B.

Marisa. K.: 2A, 2B, 2C, 3B.

L.G.‐K.: 1C, 3B.

D.K.: 1C, 3B.

P. K.‐W.: 1C, 3B.

Mariella. K.: 1C.

Maximilian. S.: 2A, 2B.

C.T.: 2A, 2B.

B.H.: 2A, 2B.

M.L.: 2A, 2B, 2C, 3B.

S.R.:3B.

D.P.: 2A, 2B, 2C, 3B.

C.E.: 3B.

S.S.: 2A, 2C, 3B.

P.S.: 1A, 1B, 1C, 2 C, 3B.

Supporting information

Data S1 Supporting Information.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.