PANDA study: subtyping of Parkinson’s disease – cohort study protocol

Abstract

Introduction

Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterised by heterogeneous motor and non-motor symptoms. The classification of PD into distinct subtypes remains a major challenge, limiting the development of precision medicine approaches. Existing subtyping methods have relied on clinical features, genetic markers or isolated biomarkers, but an integrated classification system incorporating neuroimaging and fluid biomarkers is lacking. The Prospective Assessments with Neuroimaging and fluid Data Analysis in PD (PANDA) study is a single-centre, longitudinal cohort study designed to explore PD heterogeneity by integrating multimodal neuroimaging, fluid biomarkers and comprehensive clinical assessments.

Methods and analysis

The PANDA study aims to recruit 500 PD patients and 100 healthy controls (HC) at Guangdong Provincial People’s Hospital from March 2021 to July 2026. Participants undergo three study visits: baseline, 2-year and 4-year follow-ups. Data collection includes brain MRI scans, blood and cerebrospinal fluid sampling, clinical scales, gait assessment, ambulatory blood pressure monitoring and eye-tracking evaluations. A standardised electronic data capture system ensures data accuracy and completeness. The primary objective is to identify biologically and neuroimaging-defined PD subtypes, track disease progression and explore potential biomarkers for stratification. Statistical analyses will include data-driven clustering techniques and longitudinal modelling to assess subtype stability over time.

Ethics and dissemination

The PANDA study has been approved by the Institutional Review Board of Guangdong Provincial People’s Hospital (No KY2024-719-01) and is conducted in accordance with the Declaration of Helsinki. Written informed consent is obtained from all participants or their legal guardians prior to study enrolment. The findings from this study will be disseminated through peer-reviewed journals and conference presentations, ensuring accessibility to researchers, clinicians and stakeholders involved in PD research and treatment.

Trial registration number

ChiCTR2400091709.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study integrates multimodal neuroimaging, fluid biomarkers and clinical assessments, providing a comprehensive approach to Parkinson’s disease (PD) subtyping.

• The longitudinal design allows for tracking disease progression over time, enhancing the understanding of PD heterogeneity.

• The single-centre nature of the study ensures standardised data collection, but may limit generalisability to other populations.

• The inclusion of cerebrospinal fluid biomarkers adds biological insights but is constrained by the invasive nature of lumbar puncture.

• Recruitment is ongoing, and findings will be strengthened as additional participants are enrolled and followed longitudinally.

Introduction

Parkinson’s disease (PD) is a common neurodegenerative disorder characterised by the progressive decline in both motor and non-motor functions.¹ Although traditionally viewed as a single entity, PD is in fact a heterogeneous syndrome exhibiting a wide range of phenotypic features.^{2 3} This heterogeneity is characterised by the variability of clinical presentations and disease progression, often with significant overlap between different subtypes. Without the knowledge of these subtypes, the tailor-made prognostic counselling, therapeutic interventions and personalised management would be clinically infeasible in the context of precision medicine.⁴,⁶ Clinical and demographic data have been used to define PD subtypes. For instance, PD patients have been classified into tremor-dominant and postural instability gait difficulty phenotypes based on motor features. In contrast, other subtyping systems took advantage of motor and non-motor information, such as cognition, anxiety, depression and sleep.⁷ Genetic features also play a crucial role in subtyping PD with particular focus on mutations in genes, including synuclein alpha (SNCA), leucine-rich repeat kinase 2 (LRRK2) and glucocerebrosidase (GBA) genes.⁸ However, limited studies have integrated fluid biomarkers and neuroimaging for PD subtyping.

As evidenced from prior studies, biomarkers from the blood or cerebrospinal fluid (CSF) are expected to facilitate early diagnosis and monitor disease progression in PD, emphasising their untapped potential as stratification tools for future clinical trials and other clinical studies.⁹,¹¹ The clinicopathological heterogeneity in PD patients suggests that distinct subtypes might be driven by different biological and molecular mechanisms. For instance, divergent levels of fluid biomarkers were found in each subtype.¹² Two biomarker-based staging systems have recently been proposed to classify PD patients according to biological rather than clinical characteristics, contributing to a novel framework to put into perspective the classification and staging system.^{13 14} In contrast, quantitative metrics derived from multimodality neuroimaging data have been extensively used to enhance our understanding of PD, focusing on aspects such as selective vulnerability, regional integrity of white and grey matter, pathophysiological mechanisms, differential diagnosis and disease progression.¹⁵,¹⁷ Notably, these neuroimaging metrics are not only associated with fluid biomarkers, but together they can provide complementary information for stratifying PD patients.¹⁸,²⁰ However, knowledge remains limited regarding the stratification of PD based on fluid and neuroimaging features.

We therefore launched the PANDA (Prospective Assessments with Neuroimaging and fluid Data Analysis in PD) study to prospectively investigate the subtyping of PD patients with high-quality repeated MRI scans, fluid biomarkers and comprehensive clinical assessments. This single-centre longitudinal study aims to elucidate the biological heterogeneity and brain vulnerability of PD and develop biomarkers for patient stratification in clinical research and practice. The PANDA study aims to enhance the understanding of the diversity and clinical manifestations, enabling more targeted and personalised treatment strategies in PD patients.

Methods

Study design

The PANDA study is a prospective, longitudinal, single-centre cohort study, designed to recruit 500 patients with PD and 100 healthy controls (HC). Participant enrolment began in March 2021 and is ongoing, with an expected completion by July 2026 or on reaching the recruitment target. As of December 2024, a total of 380 participants have been recruited, including 336 patients with PD and 44 HC (figure 1). Recruitment is actively continuing, with additional participants being enrolled in follow-up visits.

Figure 1. Baseline characteristics of the PANDA (Prospective Assessments with Neuroimaging and fluid Data Analysis in PD) study cohort (as of December 2024). (A) Sex distribution in the PD and HC groups. (B) Baseline age distribution in the PD and HC groups. (C) Distribution of PD patients across different subtypes (PIGD, TD and indeterminate). (D) Distribution of MDS-UPDRS scores. HC, healthy controls; MDS-UPDRS, Movement Disorder Society-Unified Parkinson’s Disease Rating Scale; PD, Parkinson’s disease; PIGD, postural instability and gait difficulty; TD, tremor-dominant.

Eligibility for participation requires that patients have a clinically established or probable PD diagnosis according to the Movement Disorders Society (MDS) criteria.²¹ Controls must be able to consent and participate and must not exhibit any signs of Parkinsonism, neurodegenerative or neuropsychiatric disorders at baseline. A detailed list of inclusion and exclusion criteria is provided in box 1. The study design is illustrated in figure 2.

Box 1. Inclusion and exclusion criteria of the PANDA study.

PD

Inclusion criteria

• Primary PD patients diagnosed according to MDS diagnostic criteria by two or more senior neurologists.

• Age: 40–80 years old, with onset of motor symptoms after age 40.

• Voluntary participation after full understanding of the study.

Exclusion criteria

• Anticholinergic drug usage.

• Clinically significant lesions observed in previous MRI or EEG scans.

• Presence of schizophrenia, delirium or hallucinations affecting clinical assessment.

• Post-DBS surgery.

• Autoimmune or infectious diseases.

HC

Inclusion criteria

• MMSE ≥26 and MoCA ≥24.

• Age: 40–80 years old.

• Ability to complete comprehensive neuropsychological assessments without self-reported cognitive decline.

• Voluntary participation with full understanding by participants and family.

• Informed consent signed by participants.

Exclusion criteria

• Abnormal cognitive function (MMSE <26 or MoCA <24).

• Clinically significant lesions observed in previous MRI or EEG scans.

• Presence of schizophrenia, delirium or hallucinations.

• Previous brain surgery or severe trauma.

• Recent use of cholinesterase inhibitors or other drugs.

• Autoimmune or infectious diseases.

• Patient or blood relative with neurodegenerative disease.

Figure 2. Overview of the PANDA study. CSF, cerebrospinal fluid; PANDA, Prospective Assessments with Neuroimaging and fluid Data Analysis in Parkinson’s disease; OCTA, optical coherence tomography angiography.

Ethical approval for initial data collection was obtained in early 2021 (No GDREC2019490H), and all participant enrolment since March 2021 has been conducted under this approval, with written informed consent obtained from all participants or their guardians. In 2024, a modified ethical approval (No KY2024-719-01) was obtained to incorporate more measurements and support study registration. The study is conducted in accordance with the Declaration of Helsinki and has been registered with the Chinese Clinical Trial Registry (ChiCTR) under the registration number ChiCTR2400091709.

Study procedures and assessments

Participants will undergo clinical assessments using both paper-based and web-based questionnaires and scales. Demographic information and medical history (ie, vascular risk factors, comorbidities), medication use (ie, drugs, dosage and duration) will be systematically collected.

Structured questionnaires and scales

The clinical assessment of PD in our study spans several dimensions, each evaluated using specific standardised instruments that have been validated and demonstrated good reliability in PD populations (table 1). Motor symptoms are primarily assessed using the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) part III, with the patients' states (‘ON’ and/or ‘OFF’) properly recorded. The MDS-UPDRS II is used to evaluate motor experiences of daily living (M-EDL), and the Wearing Off Questionnaire-9 (WOQ-9) is used to measure the wearing off phenomenon.

Table 1. Comprehensive assessments for Parkinson’s disease.

Dimension	Presentation	Instrument
Motor	Motor symptoms	MDS-UPDRS III
	Motor experiences of daily living	MDS-UPDRS II
	Wearing off	Wearing Off Questionnaire-9
Non-motor	Non-motor symptoms	Non-Motor Symptoms Scale
	Non-motor experiences of daily living	MDS-UPDRS I
	Anxiety	Hamilton Anxiety Rating Scale
	Depression	Hamilton Depression Rating Scale
	Autonomic function	Scales for Outcomes in Parkinson’s Disease—Autonomic Dysfunction
		Composite Autonomic Symptom Score
		Swallowing Disturbance Questionnaire
		Drool Rating Scale
		Wexner Constipation Questionnaire
	Sleep disorders	Pittsburgh Sleep Quality Index
		The Epworth Sleeping Scale
		Hong Kong’s development of the RBD Questionnaire
	Fatigue	Fatigue Severity Scale
Cognitive assessment	Global cognition	Mini-Mental State Examination
		Montreal Cognitive Assessment
	Attention and working memory	Symbol Digit Modalities Test
		Digital Span Test
	Executive function	Shape Trails Test
		Stroop Colour and Word Test
	Language	Boston Naming Test
		Similarities Test
	Memory	Auditory Verbal Learning Test-Huashan
		Logical Memory
	Visuospatial function	Clock Drawing Test
		Rey-Osterrieth Complex Figure
Quality of life	Quality of life	Activities of Daily Living
		Parkinson’s Disease Questionnaire-39

MDS-UPDRS, Movement Disorder Society-Unified Parkinson’s Disease Rating Scale.

Non-motor symptoms are comprehensively measured using the Non-Motor Symptoms Scale (NMSS) and the Non-Motor EDL (nM-EDL) from the MDS-UPDRS I. Anxiety and depression are assessed with the Hamilton Anxiety Rating Scale (HAM-A) and the Hamilton Depression Rating Scale (HAM-D), respectively. Autonomic function is evaluated using multiple scales, including the Scales for Outcomes in Parkinson’s Disease Autonomic Dysfunction (SCOPA-AUT), Composite Autonomic Symptom Score (COMPASS-31), Swallowing Disturbance Questionnaire, Drool Rating Scale and the Wexner Constipation Questionnaire. Sleep disorders are assessed with the Pittsburgh Sleep Quality Index (PSQI), the Epworth Sleepiness Scale (ESS) and the RBD Questionnaire. Fatigue is measured using the Fatigue Severity Scale. Quality of life is evaluated using the Activities of Daily Living scale and the Parkinson’s Disease Questionnaire-39.

Both global and domain-specific cognition will be assessed. Global cognition is evaluated with the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA); attention and working memory are measured using the Digital Span Test and Symbol Digit Modalities Test; executive function is assessed with the Shape Trails Test and Stroop Colour and Word Test; language abilities are measured using the Boston Naming Test and Similarities Test; memory is assessed with the Auditory Verbal Learning Test-Huashan and Logical Memory; visuospatial function is evaluated with the Clock Drawing Test and Rey-Osterrieth Complex Figure.

Reliability of the outcome measures

For motor assessments, the MDS-UPDRS parts II and III are the gold-standard instruments endorsed by the MDS and have undergone rigorous clinimetric testing across diverse PD cohorts, demonstrating excellent inter-rater reliability and sensitivity to clinical change.²² The WOQ-9 is also a widely used and validated tool for assessing motor fluctuations in PD patients, with strong test–retest reliability and clinical utility.²³

For non-motor symptoms, validated instruments such as the NMSS, MDS-UPDRS Part I (nM-EDL),²² HAM-A, HAM-D, SCOPA-AUT, COMPASS-31, PSQI, ESS and other related scales have been widely used in PD, showing good internal consistency, test–retest reliability and responsiveness to change.²⁴

For cognitive assessments, we follow the MDS Task Force diagnostic criteria for mild cognitive impairment in PD, incorporating validated tools such as the MoCA and MMSE for global cognition, and domain-specific tests with established use and psychometric support in PD populations.²⁵

Neuroimaging

Magnetic Resonance Imaging (MRI)

MRI scans will be used to assess both microstructural and macrostructural changes, as well as functional alterations in patients with PD. Imaging will be conducted using a Philips Ingenia 3.0 T CX MRI system equipped with a 32-channel head coil. The standardised acquisition protocol includes 3D T1-weighted imaging (T1WI), T2-weighted fluid-attenuated inversion recovery (T2 FLAIR), diffusion-WI (DWI), susceptibility-WI (SWI) and resting-state functional MRI (rs-fMRI).

(1) 3D T1-weighted image, repetition time (TR)/echo time (TE)=7.7/3.6 ms, 1 mm isotropic, slice thickness 1.0 mm, field of view (FOV)=256×256 mm, slice gap=0 mm, 192 slices. (2) SWI, 3D T1-weighted gradient-echo sequence, reconstructed voxel size=0.3×0.3×1 mm³, TR/TE/delta TE=31/7.2/6.2 ms, slice thickness 2 mm, flip angle=17°, FOV=230×190 mm, 4 echoes; (3) T2-weighted FLAIR, TR/TE/TI=9000/160/2600 ms, voxel size=0.45×0.45×5 mm³, 26 slice with three packages, slice thickness=5 mm, slice gap 1 mm; (4) DWI, using a single-shot spin echo-echo planar imaging (SE-EPI) sequence with MultiBand SENSE acceleration (MB factor=3), TR/TE 3600 ms/76 ms, flip angle=90°, 2 mm isotropic, no slice gap, 128 diffusion-weighted directions (b=1000 s/mm²) and 1 b0 image, FOV=224×224 mm; (5) rs-fMRI, using a single-shot gradient-echo EPI sequence (MB factor=6), TR/TE 800/30 ms, 2.4 mm isotropic, flip angle=52°, no slice gap, 54 slices across 380 time points, FOV=220×220 mm.

Optical coherence tomography angiography (OCTA)

OCTA images were acquired in a dark room using the AngioVue OCTA system (RTVue-XR Avanti; Optovue, Fremont, CA, USA, V.2016.2.035). The split-spectrum amplitude-decorrelation angiography software algorithm was used to evaluate vessel density. This system employs a 70 kHz spectral optical coherence tomography (OCT) platform that captures over 200 A-scans per cross-sectional B-scan, making it highly sensitive to slow capillary-level blood flow, with flow speeds estimated between 0.4 and 3 mm/s.

Two sets of imaging data were collected per assessment, each consisting of two raster volumetric patterns (one vertical and one horizontal) covering a 6×6 mm area in high-definition mode. A motion-correction algorithm was applied to merge the 3D OCT angiograms using orthogonal registration, minimising motion artefacts. Each volumetric scan included 400 scan lines. Vessel density was calculated as the percentage of signal-positive pixels relative to the total pixel count within the area of interest. The superficial and deep retinal vascular networks were generated using an automated algorithm.

The boundaries for each retinal layer were defined as follows: the superficial retinal capillary plexus extended from 3 to 15 µm from the internal limiting membrane, while the deep retinal capillary plexus extended from 15 to 70 µm below the internal limiting membrane. OCTA images with a FOV of 4.5×4.5 mm were analysed for radial peripapillary capillaries (RPC). The thicknesses of the ganglion cell complex and the retinal nerve fibre layer were also measured.

Biospecimen

Fasting blood samples will be collected, including whole blood with EDTA, EDTA plasma for cellular analysis, genomics and proteomics research, and heparin plasma for metabolomics analysis.

Due to the invasive nature of lumbar puncture, CSF samples will also be collected as many as possible. Future plans include measuring β-amyloid, total tau protein and phosphorylated tau protein (p-tau), as well as conducting α-synuclein (α-syn) seed amplification assays (SAA) to detect the presence of misfolded α-syn.

These samples will be used to examine the role of inflammation, lipid markers and other biochemical markers in PD patients. All biospecimens are appropriately stored in the Guangdong Provincial People’s Hospital Biobank, with an efficient and valid tracking system and standardised storage conditions to ensure the integrity and quality of the collected samples.

Digital biomarkers

Eye movement

Eye tracking technology has emerged as a valuable tool in the study and diagnosis of neurological disorders.²⁶ This technology measures and analyses eye movements, providing insights into cognitive processes and neurological functions.²⁷ The precise and objective data obtained from eye tracking can aid in early diagnosis, monitoring disease progression and evaluating treatment efficacy.²⁸

The head-mounted EyeKnow intelligent eye movement analysis and evaluation system (Beijing CAS-Ruiyi Information Technology Co) was used to record eye movement parameters at a sampling rate of 120 Hz. The EyeKnow system includes a tablet computer connected to a Virtual Reality (VR) headset with an eye-tracking camera. The VR headset was used to direct eye movements and capture eye movement data. PD patients were positioned in a tranquil room and equipped with the data collection apparatus on their heads. Once the patient was properly seated, a visual prompt was displayed on a screen to direct their line of sight. A calibration process involving nine points was conducted before the commencement of the study. The apparatus uses infrared imaging to discern the pupil’s relative location, reflecting eye activity through shifts in the pupil’s position. The integrated software then converts the infrared imagery into digital data, capturing both spatial and temporal information.

Ambulatory blood pressure monitoring

Ambulatory blood pressure monitoring is conducted using the DMS-ABP2 monitor (Beijing Dim Company), a portable device designed for 24-hour ambulatory blood pressure recording. The cuff is positioned on the non-dominant upper arm, 2–3 cm above the antecubital fossa, with the sensor aligned over the brachial artery pulse. Patients are instructed to continue their daily activities while avoiding excessive excitement or fatigue. Blood pressure is recorded every 30 min during the day (8:00 to 22:00) and every 60 min at night (22:00 to 8:00). For readings to be considered valid, at least 70% of the total measurements must be successful, with a minimum of 20 valid daytime readings and seven valid nighttime readings.

Gait assessment

Gait parameters were measured using the ReadyGo quantitative motor assessment system (Beijing CAS-Ruiyi Information Technology Co; figure 3). This system employs a single set of cameras, including an RGB (red/green/blue) camera and a depth camera, to capture 3D motion, using deep learning for skeletal point positioning. The ReadyGo system requires minimal setup, occupying less than 1 m² and operating effectively within a 1×5 m² area in front of the camera, without any obstructions. This device employs advanced visual perception to observe and estimate 3D joints and landmarks, track multiple skeletons simultaneously and operate without the need for subjects to wear sensors. During the gait test, participants walked at their normal pace on a 3-metre walkway without the use of mobility aids, performing one practice trial before the actual test. Final parameters, including gait speed, step length, cadence and variability, were averaged from three trials.²⁹

Figure 3. Illustrative flowchart for automated quantification of gait parameters using the ReadyGo system.

Urinary function

Monitoring and managing urine flow rate in PD is crucial for improving patients' quality of life and preventing complications such as urinary tract infections and bladder dysfunction. Urinary function is assessed using an intelligent uroflowmetry device (Andromeda, Germany). Prior to the test, subjects are instructed to drink 500–1000 mL of water and wait until they have a strong urge to urinate. No sedative medications are administered during the test. Patients urinate in their usual position, directing the urine stream towards the flow metre. Following the International Continence Society standards,³⁰ the parameters recorded include maximum urine flow rate, average urine flow rate, voided volume, actual voiding time, total voiding time, time to maximum urine flow rate, urine flow acceleration and voiding delay time. After the bladder is emptied, postvoid residual volume is measured.

Study visits

Participants will be requested to complete three study visits, including the baseline visit and two follow-up visits at years 2 and 4 (table 2). For the baseline visit, participants undergo comprehensive assessments including rating scales, blood sampling, CSF sampling, MRI and OCTA scans, gait assessment, urinary function tests, ambulatory blood pressure monitoring and eye movement assessments. At the 2-year follow-up, participants repeated rating scales, blood sampling, MRI scans, gait assessment, urinary function tests and ambulatory blood pressure monitoring. At the 4-year follow-up, the same assessments as the 2-year follow-up are conducted (table 2). Data collection for each visit takes a full day and can be extended over 2 days if preferred by the patient. All data are gathered at Guangdong Provincial People’s Hospital through face-to-face interviews and examinations conducted by centrally trained personnel, adhering to a standard protocol established by the steering committee.

Table 2. Overview of measurements and scales in the PANDA study.

	Baseline	2-year follow-up	4-year follow-up
Rating scales (see table 1)	×	×	×
Blood samples	×	×	×
CSF samples	×
MRI scans	×	×	×
OCTA scans	×
Gait assessment	×	×	×
Urinary function	×	×	×
Ambulatory blood pressure	×	×	×
Eye movement	×

CSF, cerebrospinal fluid; MRI, magnetic resonance imaging; OCTA, optical coherence tomography angiography; PANDA, Prospective Assessments with Neuroimaging and fluid Data Analysis in PD.

Outcome measures

1. Primary outcomes: identification and characterisation of PD subtypes based on integrated multimodal data, including clinical assessments, neuroimaging metrics and fluid biomarkers.

2. Secondary outcomes:

   1. Longitudinal clinical progression of motor, non-motor and cognitive assessments.

   2. Changes in fluid and imaging biomarkers over time.

   3. Quality of life and functional outcomes among subtypes.

Data collection and storage

The PANDA study follows a standardised protocol for data collection and storage to ensure the integrity, accuracy and long-term accessibility of study data. All participant data, including clinical assessments, neuroimaging, fluid biomarkers and digital biomarkers, are collected through a structured electronic data capture (EDC) system, which facilitates real-time data entry, automated quality checks and secure storage. Standardised data collection protocols and centralised training ensure data quality and consistency across study visits. Data completeness and accuracy are verified through routine quality control checks and automated validation rules within the EDC system, while data encryption and access control measures are implemented to protect participant confidentiality. Biological samples, including blood and CSF, are stored in a temperature-controlled biobank to ensure long-term preservation for future analysis. All collected data are regularly backed up on secure servers, with restricted access granted only to authorised study personnel. This structured approach to data collection and storage enhances the reproducibility and reliability of study findings.

Sample size consideration

Given the exploratory nature of the PANDA study, which seeks to uncover novel subtypes and biomarker patterns in PD, a precise formal sample size calculation is challenging due to the lack of prior data on effect sizes. Instead, we based our sample size on the number of eligible patients referred to Guangdong Provincial People’s Hospital. The study aims to recruit approximately 500 PD patients and 100 HC. This sample size was chosen to account for the variability in PD clinical manifestations and progression rates, ensuring adequate representation and generalisability across different stages and subtypes of the disease. The recruitment target also considers practical constraints, including participant availability, data collection and analysis resources, and the study’s duration. Continuous monitoring of recruitment and interim analyses will be conducted to maintain adequate power and ensure the study meets its objectives. This approach will help in validating the exploratory findings and adapting the study as necessary.

Statistical analysis

Analytical strategy for subtyping

To define PD subtypes, we plan to use data-driven, unsupervised clustering approaches that integrate multimodal data, including clinical assessments, neuroimaging metrics and fluid biomarkers. Specifically, hierarchical clustering and the latent class model will be employed to identify homogeneous patient subgroups based on multidimensional features. Additionally, dimensionality reduction techniques such as principal component analysis may be applied to reduce data complexity and enhance interpretability. The robustness and stability of the identified subtypes will be evaluated using internal validation, and external validation may be explored if available.

Handling of missing data

Missing data will be addressed based on the type and proportion of missingness. For biomarkers collected longitudinally, missing values with a low proportion and under the assumption of missing at random (MAR) will be handled using multiple imputation. If the proportion of missingness is high or suspected to be not at random, these variables may be excluded from certain analyses or handled using complete-case analyses in sensitivity analyses.³¹

For outcome data with a low proportion of missingness (eg, <5%), multiple imputation methods will also be used under the MAR assumption. If the proportion of missing outcome data exceeds 5%, sensitivity analyses, including complete-case analyses and, where appropriate, pattern-mixture models, will be performed to evaluate the potential impact of missingness on study findings.³¹

Discussion

The PANDA study aims to integrate longitudinal clinical, imaging and fluid biomarkers to identify biological and neuroimaging subtypes of PD, providing a comprehensive understanding of its pathophysiology. By incorporating multidimensional data, including neuroimaging and fluid biomarkers, this study will facilitate the characterisation of emerging clinical subtypes and contribute to a refined PD subtyping system.

Unlike other cohorts that have demonstrated their merits,³²,³⁴ the PANDA study uniquely focuses on the Chinese PD population, integrating neuroimaging and fluid biomarkers to enhance patient stratification and track disease progression. This approach aligns with recommendations for multidimensional biomarker integration, offering deeper insights into PD heterogeneity. As a single-centre longitudinal study with comprehensive follow-ups, PANDA has the potential to identify novel biomarkers and inform personalised treatment strategies.

Recent efforts to categorise and stage PD across its disease course have generated considerable discussion.^{13 14} While promising, these proposals, as noted by the MDS, require rigorous testing and validation before influencing the official disease definition, classification or staging.³⁵ Our study aims to contribute to this effort by incorporating fluid and imaging biomarkers to provide additional insights into PD classification and staging.

The biological definitions in these proposed staging frameworks rely on the presence of α-syn SAA positivity in CSF. However, it is important to note that SAA detects the presence of abnormal α-syn through laboratory amplification at supraphysiological levels, which does not necessarily reflect α-syn seeding or aggregation in the patient’s brain or neurons. Currently, most SAAs report a dichotomous positive or negative readout. While additional quantitative metrics can be calculated, providing potential quantification, relationships with motor severity are not observed.^{36 37} Additionally, there is no established method to measure α-syn biological changes over time with SAA. Other techniques can also document the presence of α-syn in different body fluids.³⁸

Given the invasiveness of CSF collection and challenges in repeat sampling, there is growing interest in identifying biomarkers in more accessible biological samples, such as blood. The measurement of brain-derived extracellular vesicle proteins in blood presents a promising approach for PD diagnosis, as these vesicles can cross the blood–brain barrier and transport neuronal contents to the periphery for easier detection.³⁹,⁴² While α-syn remains a key biomarker candidate, the multifactorial and heterogeneous nature of PD suggests the need for a biomarker panel. Additionally, Alzheimer’s disease biomarkers may be relevant to adverse cognitive outcomes in PD, and markers of brain degeneration have been associated with motor severity.^{43 44} Given the advancements in PD blood biomarker research and the feasibility of reliable measurements, our study will incorporate these biomarker categories.

Until recently, fluid biomarkers of α-syn have shown inconsistent associations with Lewy body pathology, PD diagnosis and disease progression.^{45 46} In contrast, novel imaging markers sensitive to brain tissue composition are emerging, demonstrating stronger correlations with disease severity. This shift has directed MRI studies toward investigating brain reserve and selective vulnerability. The latter refers to the preferential degeneration of neurons predisposed to accumulating disease-associated proteins and organelles.¹⁵ Early selective vulnerability in PD-affected brain regions may significantly influence motor and non-motor circuit regulation at the structural level,⁴⁷ offering deeper insights into disease progression and manifestation.

The data-driven approach, rather than one based on predefined hypotheses, is expected to identify novel biomarkers that define patient clusters with distinct clinical trajectories or neuroimaging-based microstructural changes, potentially linked to specific omics characteristics. This method could uncover new targets for future clinical trials or assay development. Additionally, the comprehensive sample and data collection allow for the exploration of emerging mechanisms and pathways. The study’s longitudinal design facilitates biomarker validation by correlating their baseline presence with disease progression and diverse clinical patterns over time.

In conclusion, unlike classification systems based solely on clinical features, this study integrates longitudinal clinical data with advanced neuroimaging and fluid biomarker analyses to identify molecularly defined PD subtypes. This approach enhances our understanding of PD heterogeneity and pathogenesis, enabling a more refined classification of the disease. By stratifying patients into homogeneous subgroups, research efforts can be more focused, improving the signal-to-noise ratio in mechanistic and treatment studies, ultimately leading to more targeted therapies. The establishment of a comprehensive cohort database and biobank will further support future research into PD mechanisms, progression and therapeutic interventions. Through collaborative efforts and ongoing data analysis, the PANDA study aims to advance PD research and contribute to the development of personalised treatments.