Evaluating an MRI-Based Machine Learning Classifier for Parkinson’s Progression Using Real-World Clinical Measures

Evaluating an MRI-Based Machine Learning Classifier for Parkinson’s Progression Using Real-World Clinical Measures | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Evaluating an MRI-Based Machine Learning Classifier for Parkinson’s Progression Using Real-World Clinical Measures Anupa A Vijayakumari, Daniel Teixeira-Dos-Santos, Hubert H Fernandez, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7870464/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Parkinson’s disease (PD) shows marked variability in disease progression, and predicting individual trajectories remains challenging. We previously developed a structural MRI–based machine learning classifier that distinguished faster from slower motor progressors using OFF-medication Movement Disorder Society–Unified Parkinson’s Disease Rating Scale Part III (MDS-UPDRS-III) scores with 89% accuracy. As OFF assessments are rarely performed in clinical practice, we evaluated whether this classifier predicts outcomes routinely used in care. Methods: Eighty-eight early PD patients from the Parkinson’s Progression Markers Initiative were previously classified as faster (n=42) or slower (n=46) motor progressors using a support vector machine model incorporating patient-specific multivariate gray matter volumetric distance and baseline clinical features. Primary outcomes were 48-month changes (Δ) in MDS-UPDRS Part II (experiences of daily living), Schwab & England Activities of Daily Living (S&E ADL), and levodopa equivalent daily dose (LEDD). Secondary analyses examined clinically meaningful thresholds: MDS-UPDRS-II worsening (≥2.51 points), ≥10% S&E ADL decline, ≥100 mg/day/year LEDD slope, and ≥1-stage Hoehn & Yahr (HY) progression. Results: Faster progressors showed significantly greater functional decline (ΔMDS-UPDRS-II: 5.31±4.77 vs. 2.76±4.56, p=0.01; ΔS&E ADL: 9.29±8.12% vs. 4.89±6.18%, p=0.009) and higher medication requirements (LEDD: 423.45±274.16 vs. 278.37±203.84 mg/day, p=0.006). Clinically meaningful deterioration was more frequent among faster progressors for MDS-UPDRS-II (81% vs. 52%, OR=3.90, p=0.009) and HY staging (62% vs. 28%, OR=4.13, p=0.003). Conclusions: An MRI-based classifier trained on OFF-medication motor assessments successfully predicts clinically meaningful deterioration across multiple real-world outcomes, supporting its potential utility for prognostic stratification in early PD. Artificial Intelligence and Machine Learning Parkinson’s disease Artificial Intelligence Machine learning MRI Parkinson’s prediction Neuroimaging Figures Figure 1 Figure 2 Introduction Parkinson's disease (PD) exhibits substantial heterogeneity in motor symptom progression, with some patients experiencing rapid decline while others maintain relatively stable motor function for years[ 10 , 30 , 43 ]. This variability poses significant challenges for clinical management, as early identification of patients at risk for faster progression could inform treatment decisions, patient counseling, and clinical trial stratification[ 1 , 17 ]. Traditional approaches to predicting disease progression have relied primarily on clinical assessment scales administered by clinicians of varying experience, which are subjective and may not fully capture the complexity of the neurobiological basis behind disease progression[ 15 , 20 ]. To address this limitation, we recently developed a structural MRI-based machine learning classifier[ 43 ] that distinguished between faster and slower motor progressors in early PD with 89% accuracy, trained on a high-quality, well-structured longitudinal research dataset from the Parkinson’s Progression Markers Initiative (PPMI)[ 23 ]. This model utilized a patient-specific multivariate gray matter volumetric distance (M GMV ) combined with baseline demographic and clinical features to predict changes in Movement Disorder Society-Sponsored Revision of the Unified Parkinson's Disease Rating Scale Part III (MDS-UPDRS-III) scores over 48 months. Importantly, the model was trained and internally validated on a held-out test set from the PPMI dataset, using OFF-medication MDS-UPDRS-III assessments that reflect true disease severity without the confounding effects of medication. For such a model to be clinically meaningful, external validation in independent, real-world cohorts is essential. However, translating this approach to routine clinical datasets presents significant practical challenges. While OFF-medication motor assessments are standard in research protocols, patients are rarely evaluated in the OFF state in routine care due to time constraints, the discomfort of medication withdrawal, and the clinical focus on symptom management rather than standardized progression measurement[ 14 , 24 ]. Given these limitations, clinicians rely on practical indicators of disease progression, including patient-reported functional outcomes and medication dosage adjustments[ 38 , 42 ]. Patient-reported outcomes such as the MDS-UPDRS Part II (experiences of daily living) and the Schwab & England Activities of Daily Living (S&E ADL) scale are particularly valuable, as they capture functional decline from the patient’s perspective and are increasingly recognized as critical components of disease monitoring[ 13 , 25 ]. Moreover, as motor symptoms worsen, clinicians typically escalate medication doses. Consequently, longitudinal changes in medication requirements can serve as an indirect but practical indicator of motor progression[ 16 ]. These changes are typically quantified using the Levodopa Equivalent Daily Dose (LEDD), a standardized metric that consolidates all antiparkinsonian medications into a single comparable value[ 42 ]. Although our prior work demonstrated MRI-based progression prediction using OFF-medication UPDRS-III scores, its applicability to routine clinical outcomes remains unclear. To address this gap, the present study examined whether patients stratified by our MRI-based classifier show differences in outcomes available within the PPMI cohort, specifically patient-reported functional measures and medication requirements. We hypothesized that individuals identified as ‘faster progressors’ by the imaging model would experience greater loss of functional independence and require larger increases in medication dosage than those identified as ‘slower progressors.’ In addition, we conducted secondary analyses to determine whether MRI-based classifications aligned with complementary indicators of disease worsening, including thresholds commonly regarded as clinically meaningful and a widely used global staging scale. Methods This study utilized the same cohort as our previously published MRI-based machine learning classifier for predicting motor progression in PD. Comprehensive descriptions of the following methods have been reported previously[43], but key details are summarized here for clarity. Participants Data were obtained from the PPMI database[23] (https://www.ppmi-info.org/access-data-specimens/download-data; RRID: SCR_006431) following standard application procedures. For current study details, please visit www.ppmi-info.org. The original study included 88 patients with early PD, selected according to inclusion and exclusion criteria described in our prior publication. Briefly, PD participants met the following inclusion criteria: recent diagnosis of PD with a positive DaTscan confirming the diagnosis; drug-naïve or no PD medication use at enrollment; and availability of baseline MRI data as well as MDS-UPDRS-III scores at baseline and at 48 months, with 48-month assessments performed in the OFF-medication state to minimize confounding from dopaminergic therapy. Exclusion criteria for all participants were a diagnosis of dementia or atypical parkinsonian syndromes, significant neurological or psychiatric conditions, and MRI artifacts or quality issues, including motion artifacts, field distortions, intensity inhomogeneities, or evidence of brain injury. Ethical Approval The Parkinson’s Progression Markers Initiative (PPMI) study complies with the principles of the Declaration of Helsinki and Good Clinical Practice and is registered on ClinicalTrials.gov (NCT01141023). All participants provided written informed consent, and study protocols were approved by the local institutional review boards or ethics committees at each participating site (see list at https://www.ppmi-info.org/about-ppmi/ppmi-clinical-sites). One of the authors (AAV) obtained permission to use tier 3 PPMI data for this study, and all MRI as well as clinical data used for this study were downloaded in a de-identified format. The PPMI Data and Publications Committee reviewed our manuscript for administrative approval in accordance with PPMI policies. Therefore, the analyses presented in this article were conducted in accordance with approved PPMI guidelines. MRI Data Acquisition and Processing T1-weighted MRI images were acquired on 3.0T scanners from multiple vendors (Siemens, Philips, and GE) using standardized PPMI acquisition parameters (see MRI operations manual: ppmi-info.org/wp-content/uploads/2017/06/PPMI-MRI-Operations-Manual-V7.pdf). Image preprocessing was performed using the Computational Anatomy Toolbox 12 (version 12.8) within Statistical Parametric Mapping (SPM12), including bias field correction, tissue segmentation into gray matter, white matter, and cerebrospinal fluid, normalization to Montreal Neurological Institute space using the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra algorithm, and smoothing with an 8mm full width at half maximum Gaussian kernel[2]. Gray matter volumes (GMV) were extracted from 40 motor-specific regions of interest using the automated anatomical labeling atlas 3 (AAL3) and normalized by total intracranial volume[36]. Inter-scanner variability was addressed using NeuroHarmonize software with diagnosis, age, and sex as biological variables[32]. Patient-Specific Multivariate Gray Matter Volumetric Distance The key innovation of our previous work was the development of a patient-specific summary score capturing GMV heterogeneity across multiple brain regions using Mahalanobis distance (M GMV )[43]. This multivariate approach calculated each patient's distance from the healthy control distribution across all 40 motor-relevant brain regions simultaneously. Age was regressed as a covariate, and the normative reference was constructed from 120 age- and sex-matched healthy controls from PPMI with no history of neurological or psychiatric illness. M GMV was then computed as where s represents the vector of age-corrected GMV observations for a single patient, μ is the vector of mean GMV values from healthy controls, and C is the covariance matrix between brain regions across healthy controls. To ensure robust estimation, M GMV was calculated relative to 1000 permutations of 100 randomly selected healthy controls, with median values reported. Machine Learning Classification Model A support vector machine classifier with a radial basis function kernel was developed using M GMV , age, sex, and baseline MDS-UPDRS-III scores as input features. The model was trained to distinguish between "Slower Progressors" and "Faster Progressors" based on changes in OFF-medication MDS-UPDRS-III scores from baseline to 48 months. Patients were categorized using the median change score (Δ = 9 points) as the threshold: slow progressors (n=46) had Δ(MDS-UPDRS-III) ≤ 9, while fast progressors (n=42) had Δ(MDS-UPDRS-III) > 9. The classifier achieved 89% accuracy with 87.5% sensitivity and 90.9% specificity on held-out test data (30% of the cohort), as detailed in our previous publication[43]. The Scikit-learn package of Python was used to develop the classifier[31]. It is important to note that the present study did not involve retraining or modifying the machine learning model. The classifier parameters were fixed from our prior publication, and all analyses applied the pre-established model labels (“faster” vs “slower” progressors) to examine their relationship with new, independent clinical outcomes that were not used during model training. Therefore, the present work represents a secondary, hypothesis-driven internal validation of clinical translation rather than model redevelopment. Clinical Outcome Measures To assess the clinical relevance of the MRI-based classification, we extracted additional progression markers from the same PPMI dataset, focusing on measures commonly available in routine practice. All measures were obtained at baseline and at the 48-month follow-up visit. Primary Outcome Measures MDS-UPDRS Part II (Experiences of Daily Living): This patient-reported outcome measures the motor-related impact of PD on daily activities across 13 items, scored from 0 (normal) to 52 (severe disability). Change scores from baseline to 48 months (ΔMDS-UPDRS-II = MDS-UPDRS-II 48M − MDS-UPDRS-II baseline ) were calculated to quantify functional decline from the patient’s perspective; positive values indicate decline[35]. Schwab & England Activities of Daily Living Scale ( S&E ADL) : The S&E ADL scale is a validated, widely recommended tool for assessing functional disability in PD. It rates overall functional capacity as a percentage of normal function (100% = completely independent; 0% = bedridden). Scores were recorded at baseline and at 48 months, with change calculated as Δ S&E ADL = S&E ADL 48M − S&E ADL baseline . Negative values indicate functional decline[39]. Levodopa Equivalent Daily Dose (LEDD): The LEDD represents the cumulative dose of all PD medications converted to levodopa equivalents, providing a standardized measure of pharmacological treatment intensity[42]. As all patients were drug-naïve at baseline (LEDD = 0), LEDD values at the 48-month visit were used to assess cumulative medication burden, reflecting disease severity and treatment response over the study period. Secondary Outcome Measures In addition to the continuous measures above, we conducted analyses using clinically meaningful thresholds to evaluate whether the MRI-based classifier could predict categorical changes that are more directly interpretable in a clinical setting. For MDS-UPDRS-II , we used minimally clinically important difference thresholds from Horváth et al. (2017). A change of ≥2.51 points was classified as worsened, while <2.51 points indicated improved or stable[13]. For S&E ADL , we applied a 10% decline threshold as a clinically significant cutoff, since one 10-point decrement represents the minimum clinically meaningful difference on this scale[6, 34, 40]. For LEDD , we calculated an annualized slope (LEDD at 48 months ÷ 4, expressed as mg/day/year; all participants were drug-naïve at baseline). We applied a threshold of ≥100 mg/day/year to define patients with substantially increased medication needs, based on prior reports[7, 44]. To enhance clinical interpretability, we additionally included OFF-medication Hoehn & Yahr (HY) staging, a global clinician-rated marker of motor severity, to test whether machine learning classifications aligned with a universally recognized PD staging system. HY classifies PD severity into stages 1–5, based on motor symptoms and functional disability. To capture meaningful worsening, we defined progression as an increase of ≥1 stage over 48 months (HY 48M – HY baseline ≥ 1). A one-stage shift is generally considered a relevant marker of advancing motor disability[9]. We also included an exploratory analysis to compare median HY stages at 48 months between progression groups. Statistical Analysis Baseline characteristics were compared between slower and faster progressors using independent t-tests for continuous variables, chi-square tests for categorical variables, and Mann–Whitney U tests for ordinal variables (HY stage). Primary analyses focused on three clinical outcomes at 48 months: LEDD change in MDS-UPDRS-II scores, and change in Schwab & England ADL scores. Effect sizes were calculated using Cohen’s d[8]. Secondary analyses evaluated whether our progression classification aligned with established definitions of clinically meaningful change by examining the proportion of patients meeting published thresholds or cutoffs, with group differences tested using chi-square. Odds ratios (ORs) with 95% confidence intervals were reported to quantify the strength of association between progression group and clinically meaningful outcomes. All analyses were conducted in Python (v3.12.11) using pandas, NumPy, SciPy, statsmodels, and matplotlib, with significance set at p < 0.05 after Benjamini–Hochberg correction for multiple comparisons[4]. Results Baseline Characteristics As shown in Table 1, there were no significant differences in age (60.43 ± 9.81 vs. 59.75 ± 9.05 years, p = 0.66), sex distribution (78.6% vs. 58.7% male, p = 0.06), MDS-UPDRS-II scores (4.69 ± 3.13 vs. 5.74 ± 4.57, p = 0.16), S&E ADL scores (95.24% ± 4.93 vs. 94.35% ± 5.01, p = 0.40), or HY stage (median [IQR]: 1 [1–2] vs. 2 [1–2], p = 0.14). However, slower progressors demonstrated significantly higher MDS-UPDRS-III motor scores at baseline (22.41 ± 9.04 vs. 16.71 ± 7.55, p = 0.002). Table 1 Baseline Characteristics by MRI-Based Progression Classification Characteristic Faster Progressors (n = 42) Slower Progressors (n = 46) p -value Age, years (mean ± SD) 60.43 ± 9.81 59.75 ± 9.05 0.66 Sex, male n (%) 33 (78.6) 27 (58.7) 0.06 MDS-UPDRS-II baseline 4.69 ± 3.13 5.74 ± 4.57 0.16 MDS-UPDRS-III baseline 16.71 ± 7.55 22.41 ± 9.04 0.002 S&E ADL baseline, % 95.24 ± 4.93 94.35 ± 5.01 0.40 HY baseline, median (IQR) 1 (1–2) 2 (1–2) 0.14 Values are mean ± standard deviation (SD) except for HY, where a non-parametric test was employed. n (%), or median (interquartile range, IQR) as appropriate. Abbreviations: MDS-UPDRS-II = Movement Disorder Society–sponsored revision of the Unified Parkinson’s Disease Rating Scale Part II; MDS-UPDRS-III = Movement Disorder Society–sponsored revision of the Unified Parkinson’s Disease Rating Scale Part III; S&E ADL = Schwab & England Activities of Daily Living; HY = Hoehn & Yahr. Clinical Progression at 48 Months—Primary Analyses At 48 months, faster progressors demonstrated significantly greater clinical deterioration across all primary outcome measures compared to slower progressors (Table 2). The mean increase in MDS-UPDRS-II scores was 5.31 ± 4.77 for faster progressors versus 2.76 ± 4.56 for slower progressors ( p = 0.01, Cohen's d = 0.54). S&E ADL independence declined by 9.29 ± 8.12% in faster progressors compared to 4.89 ± 6.18% in slower progressors ( p = 0.009, Cohen's d = 0.57). LEDD requirements at 48 months were 423.45 ± 274.16 mg/day for faster progressors versus 278.37 ± 203.84 mg/day for slower progressors ( p = 0.006, Cohen's d = 0.60). All effect sizes indicated moderate to large clinically meaningful differences between groups. Table 2 Clinical Progression Markers at 48 Months by MRI-Based Classification Outcome Measure Faster Progressors (n=42) Slower Progressors (n=46) p -value Cohen's d Δ MDS-UPDRS-II 5.31 ± 4.77 2.76 ± 4.56 0.01 0.54 Δ S&E ADL, % −9.29 ± 8.12 −4.89 ± 6.18 0.009 0.57 LEDD at 48M, mg/day 423.45 ± 274.16 278.37 ± 203.84 0.006 0.60 Data presented as mean ± standard deviation. Δ indicates change from baseline to 48 months. Abbreviations: MDS-UPDRS-II = Movement Disorder Society–sponsored revision of the Unified Parkinson’s Disease Rating Scale, Part II (Motor Aspects of Experiences of Daily Living); S&E ADL = Schwab & England Activities of Daily Living. LEDD = Levodopa Equivalent Daily Dose. Statistical significance at p < 0.05 Clinical Meaningfulness Analysis—Secondary Analyses Secondary analyses revealed distinct patterns of clinically meaningful progression between groups (Table 3; Figure 1). Faster progressors were significantly more likely to experience worsening in daily functioning, with 81% exceeding the threshold for clinically meaningful decline on the MDS-UPDRS-II compared to 52% of slower progressors (OR = 3.90, 95% CI 1.49–10.21, p = 0.009). A trend was observed for activities of daily living, where 62% of faster progressors showed ≥10% decline on the S&E ADL scale versus 41% of slower progressors (OR = 2.31, 95% CI 0.98–5.43, p = 0.05). With respect to medication use, 48% of faster progressors exceeded the annualized LEDD slope threshold of ≥100 mg/day/year compared to 30% of slower progressors (OR = 2.08, 95% CI 0.87–4.97, p = 0.09), which was not statistically significant. The most pronounced difference was observed in HY staging, where 62% of faster progressors experienced ≥1 stage worsening over 48 months, compared to only 28% of slower progressors (OR = 4.13, 95% CI 1.69–10.09, p = 0.003; Figure 2). Supporting this finding, exploratory analysis revealed that faster progressors had higher HY stages at 48 months (median [IQR] 2 [2–2]) compared to slower progressors (2 [1–2], p = 0.012). The distributional shifts in HY stages between faster and slower progressors are provided in Supplementary File 1. Table 3 Clinically Meaningful Progression Thresholds at 48 Months by MRI-Based Classification. Outcome Measure Thresholds Faster Progressors (n=42) Slower Progressors (n=46) Odds ratio (95% CI) p -value MDS-UPDRS-II Δ < 2.51 Δ ≥ 2.51 8 (19%) 34 (81%) 22 (48%) 24 (52%) 3.90 (1.49–10.21) 0.009 S&E ADL Δ < 10% Δ ≥ 10% 16 (38%) 26 (62%) 27 (59%) 19 (41%) 2.31 (0.98–5.43) 0.05 LEDD slope < 100 mg/day/year ≥ 100 mg/day/year 22 (52%) 20 (48%) 32 (70%) 14 (30%) 2.08 (0.87–4.97) 0.09 HY Δ < 1 Δ ≥ 1 16 (38%) 26 (62%) 33 (72%) 13 (28%) 4.13 (1.69–10.09) 0.003 Values are n (%). Δ = change from baseline to 48 months. Abbreviations: CI = Confidence Interval; MDS-UPDRS-II = Movement Disorder Society–sponsored revision of the Unified Parkinson’s Disease Rating Scale Part II; S&E ADL = Schwab & England Activities of Daily Living; LEDD = Levodopa Equivalent Daily Dose; HY = Hoehn & Yahr. Statistical significance at p < 0.05. Discussion In this study, we evaluated whether our previously developed machine learning classifier[43] could predict pragmatic outcomes commonly available in clinical practice. The classifier was originally trained on 48-month OFF-medication MDS-UPDRS Part III change scores to stratify patients as faster or slower motor progressors. Here, we performed an internal validation within the PPMI cohort, applying the classifier to new, clinically relevant outcomes not used in model training. Patients identified as faster progressors exhibited greater functional decline and pharmacologic burden compared to slower progressors, providing preliminary support for the potential utility of structural MRI-based prognostic models for predicting disease progression in early PD.An intriguing finding emerged when comparing baseline characteristics: slower progressors had significantly higher baseline MDS-UPDRS-III motor scores compared to faster progressors (22.41 vs. 16.71, p = 0.002). While this initially appears counterintuitive, it may reflect important underlying pathophysiological mechanisms. This paradox suggests that baseline motor severity does not necessarily reflect an individual patient’s progression trajectory, and that underlying neurobiological mechanisms captured by MRI may be more predictive of future decline. These findings point toward the existence of distinct PD subgroups, in which structural brain alterations differentially shape the relationship between clinical presentation and subsequent disease trajectory. To further evaluate the clinical relevance of these subgroups, we assessed outcomes that directly reflect patient function and independence. MDS-UPDRS-II and S&E ADL represent medication-independent assessments of functional decline, making them valuable tools for monitoring progression[11, 37, 39]. We observed that faster progressors demonstrated significantly greater deterioration in ΔMDS-UPDRS-II scores, with 81% experiencing a clinically meaningful decline compared with 52% of slower progressors. This finding was consistent with ΔS&E ADL scores, where faster progressors showed greater loss of independence (62% with ≥10% decline vs 41% of slower progressors). Although this difference reached only borderline significance ( p = 0.05), it reflected the broader pattern of functional deterioration across complementary domains. Unlike motor examinations (MDS-UPDRS-III), which require trained raters and are medication status-dependent, both MDS-UPDRS Part II and S&E ADL can be completed directly by patients, supporting their utility for routine monitoring[29]. The ability of our imaging classifier to capture decline across multiple patient-centered domains underscores its translational potential in clinical practice. In addition to functional deterioration, those classified as faster progressors required increased medication over time. At 48 months, they had significantly higher LEDD than slower progressors, consistent with prior evidence that worsening motor symptoms drive increased treatment needs[3, 21]. Supporting this pattern, Birkenbihl et al. also reported that fast-progressing patients required higher LEDD after approximately four years of follow-up[5]. Together, these findings support the notion that patients at higher risk for rapid motor progression demand greater treatment intensity over time. Importantly, greater medication use did not mitigate decline, but rather co-occurred with it, underscoring the robustness of the classifier’s predictions. Despite these promising results, our clinically meaningful threshold-based analysis revealed important limitations. Using a cutoff derived from prior literature (~100 mg/day/year)[7, 44], the non-significant trend in LEDD progression ( p = 0.09) warrants careful interpretation. While LEDD is widely considered a surrogate for disease severity[42], it is strongly influenced by prescribing practices, including the timing of levodopa initiation, choice of adjunct therapies, and responsiveness to patient complaints[18, 19, 33]. Moreover, conversion formulas are approximations that introduce variability,[16, 26] with scaling differences depending on the weight assigned to each drug class. As a result, LEDD progression may reflect both disease severity and physician prescribing patterns, which complicates its interpretation as a marker of clinical progression[27, 41]. This inherent confounding suggests caution in using medication burden as a primary endpoint for evaluating progression prediction models. Future studies should prioritize outcomes that are more directly tied to disease biology and less dependent on prescriber variability. To assess our classifier's discriminative ability across different clinical measures, we explored whether it could identify differences in HY staging. While HY is also medication-dependent like motor examinations (MDS-UPDRS-III), it represents a universally recognized staging system widely used in clinical practice, making it a valuable benchmark for our model. Notably, HY staging revealed the most pronounced group difference, with 62% of faster progressors experiencing a ≥1-stage worsening compared with only 28% of slower progressors. Although the scale is coarse and ordinal, a one-stage increase is generally regarded as a clinically meaningful indicator of symptom progression in PD[9, 28]. The ability of our model to stratify patients by their risk of HY progression highlights a key translational strength by connecting imaging-based predictions to a simple clinical tool already central to practice. We also compared HY scores at 48 months, which confirmed higher staging among faster progressors. This reinforces that HY progression differences were robust and not solely dependent on categorical thresholds. Conventionally, structural T1-weighted MRI is not considered useful for prognosis in PD, as routine scans often appear “normal”[22]. In our original work, however, we introduced a patient-specific multivariate gray matter volumetric distance (M GMV ) derived using Mahalanobis distance, which summarized atrophy in the gray matter regions across multiple regions into a single score[43]. When combined with machine learning, this approach demonstrated that even standard MRI data can extract prognostically valuable information by identifying patients at risk for faster motor deterioration. Importantly, while that model was trained and internally validated using OFF-medication MDS-UPDRS-III scores, direct external validation remains limited because OFF assessments are rarely available in clinical practice. The present findings suggest that future prognostic models need not rely exclusively on such OFF assessments. Instead, pragmatic measures such as MDS-UPDRS-II and the S&E ADL scale—more commonly collected in routine care—may serve as equally meaningful and clinically relevant endpoints. This shift would allow machine learning models to undergo both rigorous internal and external validation in real-world datasets[12]. Taken together, our work underscores that baseline T1-weighted MRI, often deemed clinically uninformative, can provide a powerful substrate for prognostic modeling when coupled with advanced analytic methods. Although these findings show promise, future development must take into account several limitations of the current study. First, while this work was conducted in the well-characterized PPMI research cohort[23], future studies should evaluate such models in clinical datasets, where real-world challenges such as variability in imaging protocols, treatment strategies, and follow-up are unavoidable. In particular, external validation in non-PPMI cohorts, including hospital-based populations with more heterogeneous data, will be critical to establish generalizability. Second, our original classifier was trained on only 88 patients, reflecting the relatively small subset of PPMI participants with complete OFF-medication MDS-UPDRS-III data. In contrast, other clinical measures such as MDS-UPDRS-II or S&E ADL are more widely available both in PPMI and in routine practice, and anchoring models to these outcomes would enable training on larger and more representative samples. Finally, the 48-month follow-up period may not capture the full course of disease progression. Extended follow-up will be critical to establish whether predictive performance persists and continues to hold clinical relevance over time. Addressing these limitations will be essential for developing MRI-based prognostic tools that are both biologically grounded and clinically applicable in everyday PD care. Given the wide spectrum of PD progression ranging from relatively benign to rapidly advancing forms, the ability to anticipate disease course carries substantial clinical and research value. Early identification of likely faster progressors could assist clinicians in tailoring follow-up intensity, optimizing therapeutic timing, and counselling patients and caregivers. Prognostic stratification also holds implications for clinical trial design—facilitating enrichment of disease-modifying trials with higher-risk participants—and may ultimately support the discovery and validation of biomarkers linked to progression. Therefore, a reliable and accessible predictor of disease course addresses a critical unmet need in PD care and research. In conclusion, this study demonstrated that an MRI-based machine learning classifier, originally trained to predict OFF-medication motor progression, successfully identified patients at risk for clinically meaningful deterioration across multiple real-world outcome measures. Patients classified as faster progressors exhibited significantly greater functional decline on patient-reported measures, higher medication requirements, and more pronounced staging progression compared to slower progressors. Together, these findings represent an important step toward establishing the clinical utility of structural MRI–based prognostic models for predicting disease progression in early PD. Such prognostic information could inform clinical trial stratification, guide patient counseling, and potentially support decisions about treatment timing. Declarations Acknowledgements Data used in the preparation of this article were obtained on 2025-07-28 from the Parkinson’s Progression Markers Initiative (PPMI) database (https://www.ppmi-info.org/access-data-specimens/download-data), RRID:SCR_006431. For up-to-date information on the study, visit http://www.ppmi-info.org. PPMI – a public-private partnership – is funded by the Michael J. Fox Foundation for Parkinson’s Research and funding partners, including 4D Pharma, Abbvie, AcureX, Allergan, Amathus Therapeutics, Aligning Science Across Parkinson's, AskBio, Avid Radiopharmaceuticals, BIAL, BioArctic, Biogen, Biohaven, BioLegend, BlueRock Therapeutics, Bristol-Myers Squibb, Calico Labs, Capsida Biotherapeutics, Celgene, Cerevel Therapeutics, Coave Therapeutics, DaCapo Brainscience, Denali, Edmond J. Safra Foundation, Eli Lilly, Gain Therapeutics, GE HealthCare, Genentech, GSK, Golub Capital, Handl Therapeutics, Insitro, Jazz Pharmaceuticals, Johnson & Johnson Innovative Medicine, Lundbeck, Merck, Meso Scale Discovery, Mission Therapeutics, Neurocrine Biosciences, Neuron23, Neuropore, Pfizer, Piramal, Prevail Therapeutics, Roche, Sanofi, Servier, Sun Pharma Advanced Research Company, Takeda, Teva, UCB, Vanqua Bio, Verily, Voyager Therapeutics, the Weston Family Foundation and Yumanity Therapeutics. Author Contributions A.A. Vijayakumari : study concept and design; data extraction; data analysis and interpretation; drafting and revision of the manuscript for content. D. Teixeira-Dos-Santos : critical revision of the manuscript for content; input on study interpretation. H.H. Fernandez : study concept and design; critical revision of the manuscript for content. B.L. Walter : study concept and design; critical revision of the manuscript for content. Data Availability The data analyzed in this study are available from the Parkinson’s Progression Markers Initiative (PPMI) database (https://www.ppmi-info.org/access-data-specimens/download-data; RRID:SCR_006431) upon application and approval by the PPMI Data and Publications Committee. They are also available from the corresponding author upon reasonable request. Conflict of interest The authors declare that they have no conflict of interest. Competing interests The authors have no competing interests to declare that are relevant to the contents of this article. Funding This work was supported by a Cleveland Clinic Neurological Institute Clinician Scientist Career Award (A.A.V.). References Armstrong MJ, Okun MS (2020) Diagnosis and Treatment of Parkinson Disease: A Review. Jama 323:548-560 Ashburner J, Friston KJ (2005) Unified segmentation. Neuroimage 26:839-851 Bartl M, Dakna M, Schade S, Wicke T, Lang E, Ebentheuer J, Weber S, Trenkwalder C, Mollenhauer B (2021) Longitudinal Change and Progression Indicators Using the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale in Two Independent Cohorts with Early Parkinson’s Disease. Journal of Parkinson’s Disease 12:437-452 Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B (Methodological) 57:289-300 Birkenbihl C, Ahmad A, Massat NJ, Raschka T, Avbersek A, Downey P, Armstrong M, Fröhlich H (2023) Artificial intelligence-based clustering and characterization of Parkinson's disease trajectories. Scientific Reports 13:2897 Bouça-Machado R, Fernandes A, Ranzato C, Beneby D, Nzwalo H, Ferreira JJ (2022) Measurement tools to assess activities of daily living in patients with Parkinson's disease: A systematic review. Front Neurosci 16:945398 Bovenzi R, Conti M, Degoli GR, Cerroni R, Simonetta C, Liguori C, Salimei C, Pisani A, Pierantozzi M, Stefani A, Mercuri NB, Schirinzi T (2023) Shaping the course of early-onset Parkinson's disease: insights from a longitudinal cohort. Neurol Sci 44:3151-3159 Cohen J (2013) Statistical power analysis for the behavioral sciences. routledge Goetz CG, Poewe W, Rascol O, Sampaio C, Stebbins GT, Counsell C, Giladi N, Holloway RG, Moore CG, Wenning GK, Yahr MD, Seidl L (2004) Movement Disorder Society Task Force report on the Hoehn and Yahr staging scale: status and recommendations. Mov Disord 19:1020-1028 Greenland JC, Williams-Gray CH, Barker RA (2019) The clinical heterogeneity of Parkinson's disease and its therapeutic implications. Eur J Neurosci 49:328-338 Harrison MB, Wylie SA, Frysinger RC, Patrie JT, Huss DS, Currie LJ, Wooten GF (2009) UPDRS activity of daily living score as a marker of Parkinson's disease progression. Mov Disord 24:224-230 Ho SY, Phua K, Wong L, Bin Goh WW (2020) Extensions of the External Validation for Checking Learned Model Interpretability and Generalizability. Patterns (N Y) 1:100129 Horváth K, Aschermann Z, Kovács M, Makkos A, Harmat M, Janszky J, Komoly S, Karádi K, Kovács N (2017) Minimal clinically important differences for the experiences of daily living parts of movement disorder society-sponsored unified Parkinson's disease rating scale. Mov Disord 32:789-793 Hoskovcová M, Dušek P, Sieger T, Brožová H, Zárubová K, Bezdíček O, Šprdlík O, Jech R, Štochl J, Roth J, Růžička E (2015) Predicting Falls in Parkinson Disease: What Is the Value of Instrumented Testing in OFF Medication State? PLoS One 10:e0139849 Jankovic J, McDermott M, Carter J, Gauthier S, Goetz C, Golbe L, Huber S, Koller W, Olanow C, Shoulson I, et al. (1990) Variable expression of Parkinson's disease: a base-line analysis of the DATATOP cohort. The Parkinson Study Group. Neurology 40:1529-1534 Jost ST, Kaldenbach MA, Antonini A, Martinez-Martin P, Timmermann L, Odin P, Katzenschlager R, Borgohain R, Fasano A, Stocchi F, Hattori N, Kukkle PL, Rodríguez-Violante M, Falup-Pecurariu C, Schade S, Petry-Schmelzer JN, Metta V, Weintraub D, Deuschl G, Espay AJ, Tan EK, Bhidayasiri R, Fung VSC, Cardoso F, Trenkwalder C, Jenner P, Ray Chaudhuri K, Dafsari HS (2023) Levodopa Dose Equivalency in Parkinson's Disease: Updated Systematic Review and Proposals. Mov Disord 38:1236-1252 Kalia LV, Lang AE (2015) Parkinson's disease. Lancet 386:896-912 Kalilani L, Friesen D, Boudiaf N, Asgharnejad M (2019) The characteristics and treatment patterns of patients with Parkinson's disease in the United States and United Kingdom: A retrospective cohort study. PLoS One 14:e0225723 Ku M, Je NK (2023) Exploring the prescribing trends and factors affecting initial anti-parkinsonian drug selection in Korea: A nationwide population-based cohort study. Journal of Clinical Neuroscience 116:60-66 Lim S-Y, Tan AH (2018) Historical perspective: The pros and cons of conventional outcome measures in Parkinson's disease. Parkinsonism & Related Disorders 46:S47-S52 Lin YH, Fang TC, Lei HB, Chiu SC, Chang MH, Guo YJ (2023) UPSIT subitems may predict motor progression in Parkinson's disease. Front Neurol 14:1265549 Mahlknecht P, Hotter A, Hussl A, Esterhammer R, Schocke M, Seppi K (2010) Significance of MRI in diagnosis and differential diagnosis of Parkinson's disease. Neurodegener Dis 7:300-318 Marek K, Chowdhury S, Siderowf A, Lasch S, Coffey CS, Caspell-Garcia C, Simuni T, Jennings D, Tanner CM, Trojanowski JQ, Shaw LM, Seibyl J, Schuff N, Singleton A, Kieburtz K, Toga AW, Mollenhauer B, Galasko D, Chahine LM, Weintraub D, Foroud T, Tosun-Turgut D, Poston K, Arnedo V, Frasier M, Sherer T (2018) The Parkinson's progression markers initiative (PPMI) - establishing a PD biomarker cohort. Ann Clin Transl Neurol 5:1460-1477 Martin H, Ali R, Sriram A, Coleman R, Ruether E, Boyce H, Maley ML, Khan M (2025) Leukoaraiosis does not impact motor outcomes in Parkinson's patients post deep brain stimulation. Clin Park Relat Disord 12:100348 Muslimovic D, Post B, Speelman JD, Schmand B, de Haan RJ (2008) Determinants of disability and quality of life in mild to moderate Parkinson disease. Neurology 70:2241-2247 Nyholm D, Jost WH (2021) An updated calculator for determining levodopa-equivalent dose. Neurological Research and Practice 3:58 Orayj K, Lane E (2019) Patterns and Determinants of Prescribing for Parkinson's Disease: A Systematic Literature Review. Parkinsons Dis 2019:9237181 Pagano G, De Micco R, Yousaf T, Wilson H, Chandra A, Politis M (2018) REM behavior disorder predicts motor progression and cognitive decline in Parkinson disease. Neurology 91:e894-e905 Parashos SA, Luo S, Biglan KM, Bodis-Wollner I, He B, Liang GS, Ross GW, Tilley BC, Shulman LM (2014) Measuring disease progression in early Parkinson disease: the National Institutes of Health Exploratory Trials in Parkinson Disease (NET-PD) experience. JAMA Neurol 71:710-716 Passaretti M, Veréb D, Mijalkov M, Chang Y-W, Zhao H, Zufiria-Gerbolés B, Sun J, Volpe G, Rivera N, Bologna M, Pereira JB (2025) Clinical progression and genetic pathways in body-first and brain-first Parkinson’s disease. Molecular Neurodegeneration 20:74 Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss R, Dubourg V, Vanderplas J, Passos A, Cournapeau D, Brucher M, Perrot M, Duchesnay É (2011) Scikit-learn: Machine learning in Python. Journal of machine Learning research 12: 2825-2830 Pomponio R, Erus G, Habes M, Doshi J, Srinivasan D, Mamourian E, Bashyam V, Nasrallah IM, Satterthwaite TD, Fan Y, Launer LJ, Masters CL, Maruff P, Zhuo C, Volzke H, Johnson SC, Fripp J, Koutsouleris N, Wolf DH, Gur R, Gur R, Morris J, Albert MS, Grabe HJ, Resnick SM, Bryan RN, Wolk DA, Shinohara RT, Shou H, Davatzikos C (2020) Harmonization of large MRI datasets for the analysis of brain imaging patterns throughout the lifespan. Neuroimage 208:116450 Pringsheim T, Day GS, Smith DB, Rae-Grant A, Licking N, Armstrong MJ, de Bie RMA, Roze E, Miyasaki JM, Hauser RA, Espay AJ, Martello JP, Gurwell JA, Billinghurst L, Sullivan K, Fitts MS, Cothros N, Hall DA, Rafferty M, Hagerbrant L, Hastings T, O'Brien MD, Silsbee H, Gronseth G, Lang AE (2021) Dopaminergic Therapy for Motor Symptoms in Early Parkinson Disease Practice Guideline Summary: A Report of the AAN Guideline Subcommittee. Neurology 97:942-957 Ramaker C, Marinus J, Stiggelbout AM, van Hilten BJ (2002) Systematic evaluation of rating scales for impairment and disability in Parkinson’s disease. Movement Disorders 17:867-876 Rodriguez-Blazquez C, Rojo-Abuin JM, Alvarez-Sanchez M, Arakaki T, Bergareche-Yarza A, Chade A, Garretto N, Gershanik O, Kurtis MM, Martinez-Castrillo JC, Mendoza-Rodriguez A, Moore HP, Rodriguez-Violante M, Singer C, Tilley BC, Huang J, Stebbins GT, Goetz CG, Martinez-Martin P (2013) The MDS-UPDRS Part II (motor experiences of daily living) resulted useful for assessment of disability in Parkinson's disease. Parkinsonism Relat Disord 19:889-893 Rolls ET, Huang CC, Lin CP, Feng J, Joliot M (2020) Automated anatomical labelling atlas 3. Neuroimage 206:116189 Sampaio C (2009) Can focusing on UPDRS Part II make assessments of Parkinson disease progression more efficient? Nature Reviews Neurology 5:130-131 Schrag A, Sampaio C, Counsell N, Poewe W (2006) Minimal clinically important change on the unified Parkinson's disease rating scale. Mov Disord 21:1200-1207 Schwab RS, England AC, Jr. (1969) Projection technique for evaluating surgery in Parkinson's disease. In: Gillingham FJ, Donaldson IML (eds) Third Symposium on Parkinson’s Disease. E & S Livingstone, Edinburgh, Scotland, pp 152-157 Siderowf A (2010) Schwab and England Activities of Daily Living Scale. In: Kompoliti K, Verhagen L (eds) Encyclopedia of Movement Disorders. Elsevier, Amsterdam, pp 99-100 Tolosa E, Ebersbach G, Ferreira JJ, Rascol O, Antonini A, Foltynie T, Gibson R, Magalhaes D, Rocha JF, Lees A (2021) The Parkinson's Real-World Impact Assessment (PRISM) Study: A European Survey of the Burden of Parkinson's Disease in Patients and their Carers. J Parkinsons Dis 11:1309-1323 Tomlinson CL, Stowe R, Patel S, Rick C, Gray R, Clarke CE (2010) Systematic review of levodopa dose equivalency reporting in Parkinson's disease. Mov Disord 25:2649-2653 Vijayakumari AA, Fernandez HH, Walter BL (2023) MRI-based multivariate gray matter volumetric distance for predicting motor symptom progression in Parkinson's disease. Scientific Reports 13:17704 Wilson J, Alcock L, Yarnall AJ, Lord S, Lawson RA, Morris R, Taylor JP, Burn DJ, Rochester L, Galna B (2020) Gait Progression Over 6 Years in Parkinson's Disease: Effects of Age, Medication, and Pathology. Front Aging Neurosci 12:577435 Additional Declarations The authors declare no competing interests. Supplementary Files SupplementaryInformation.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7870464","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":530209117,"identity":"3ad4722e-305c-45ac-b61c-1ea983189a94","order_by":0,"name":"Anupa A Vijayakumari","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAs0lEQVRIie3PrQ7CMBSG4W+pXTI7td1CFzzXUjIxwwieXzeJI7sMLgBxlibMwBXMVKGLR1AyDKqtI6Gvf/KdA4RCP1kMUhwZECkPIjgmAOPuBAKY7Z1Jcrx1JJbT6tBI4HHe2Ek6LIQ5rKzba4movfcOM8OcG8LqU5oQi+liF/lIdhVPGdwIH4kUH7K2k2L8pS/ev3QtkZ1kQy21fq7ypJGR0rS1k6/MhPQkJt+VUCgU+ode7EQ9AJgXtlAAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0001-7041-9454","institution":"Cleveland Clinic","correspondingAuthor":true,"prefix":"","firstName":"Anupa","middleName":"A","lastName":"Vijayakumari","suffix":""},{"id":530209118,"identity":"5cc713db-79d5-4076-a1a9-a608c589cedb","order_by":1,"name":"Daniel Teixeira-Dos-Santos","email":"","orcid":"","institution":"Cleveland Clinic","correspondingAuthor":false,"prefix":"","firstName":"Daniel","middleName":"","lastName":"Teixeira-Dos-Santos","suffix":""},{"id":530209119,"identity":"82aa1340-1630-4dc8-80f9-256994f5794e","order_by":2,"name":"Hubert H Fernandez","email":"","orcid":"","institution":"Cleveland Clinic","correspondingAuthor":false,"prefix":"","firstName":"Hubert","middleName":"H","lastName":"Fernandez","suffix":""},{"id":530209120,"identity":"53664e22-2f0a-464b-bbfc-adff71527619","order_by":3,"name":"Benjamin L Walter","email":"","orcid":"","institution":"Cleveland Clinic","correspondingAuthor":false,"prefix":"","firstName":"Benjamin","middleName":"L","lastName":"Walter","suffix":""}],"badges":[],"createdAt":"2025-10-15 16:57:06","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7870464/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7870464/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93779913,"identity":"8d02b37c-9598-4b1b-b27c-8d7217f87fee","added_by":"auto","created_at":"2025-10-17 12:54:48","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":124594,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/581302dcef3fc3ea6389da2e.docx"},{"id":93778381,"identity":"3009df04-fb6c-4a2c-95c2-9e66780ee6ec","added_by":"auto","created_at":"2025-10-17 12:46:48","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs7870464.json","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/3f4154e16cb999a0cfbf4ae5.json"},{"id":93778384,"identity":"edc8ce41-115e-4748-a8eb-97a505d823a1","added_by":"auto","created_at":"2025-10-17 12:46:48","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":135680,"visible":true,"origin":"","legend":"","description":"","filename":"rs78704640enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/56e8805176dab0081550a134.xml"},{"id":93778388,"identity":"1cf4c979-90db-47f8-86e4-caf0d1ada7e4","added_by":"auto","created_at":"2025-10-17 12:46:48","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":133163,"visible":true,"origin":"","legend":"","description":"","filename":"rs78704640structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/a61d1257fbf36df6113c42d8.xml"},{"id":93778387,"identity":"fe6f305e-24ca-44c7-b6ad-0c716cfc0ecf","added_by":"auto","created_at":"2025-10-17 12:46:48","extension":"html","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":142327,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/ec08fe02712f5e8befc7f4ce.html"},{"id":93779912,"identity":"88e92753-9f3e-4ffc-bfd5-729b56fd004c","added_by":"auto","created_at":"2025-10-17 12:54:48","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":33849,"visible":true,"origin":"","legend":"\u003cp\u003eForest plot of clinically meaningful outcomes at 48 months comparing faster and slower progressors. Odds ratios with 95% confidence intervals (horizontal lines) are displayed for MDS-UPDRS-II (Δ ≥ 2.51), S\u0026amp;E ADL (≥10% decline), LEDD slope (≥100 mg/day/year), and HY stage (≥1 stage worsening). The vertical dashed line represents no difference (OR = 1). The X-axis is plotted on a logarithmic scale.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/f988fa33c8b0b4296d52756a.png"},{"id":93778383,"identity":"bb66cdf2-22e1-4408-bd77-df04d96ce8ac","added_by":"auto","created_at":"2025-10-17 12:46:48","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":34115,"visible":true,"origin":"","legend":"\u003cp\u003eStacked bar plots illustrate the proportion of patients with and without clinically meaningful worsening (≥1 stage increase in Hoehn \u0026amp; Yahr [HY] staging) over 48 months. Bars represent faster (n = 42) and slower (n = 46) MRI-based progressors. The teal segments indicate patients without HY progression (Δ \u0026lt; 1), and the pink segments indicate patients with ≥1 stage progression (Δ ≥ 1).\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/148bdc4bdd8db294f8b8e351.png"},{"id":93780277,"identity":"e8b990fc-3bf8-4a9a-9dfa-723b5d829860","added_by":"auto","created_at":"2025-10-17 13:02:48","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":970124,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/62dbe759-bc32-4340-b7a8-ca18e7d80564.pdf"},{"id":93778385,"identity":"062c10f2-7f2d-4fef-9e59-50cae29edc40","added_by":"auto","created_at":"2025-10-17 12:46:48","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":137329,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7870464/v1/0480e40e151b86391dc772f5.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eEvaluating an MRI-Based Machine Learning Classifier for Parkinson’s Progression Using Real-World Clinical Measures\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eParkinson's disease (PD) exhibits substantial heterogeneity in motor symptom progression, with some patients experiencing rapid decline while others maintain relatively stable motor function for years[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. This variability poses significant challenges for clinical management, as early identification of patients at risk for faster progression could inform treatment decisions, patient counseling, and clinical trial stratification[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Traditional approaches to predicting disease progression have relied primarily on clinical assessment scales administered by clinicians of varying experience, which are subjective and may not fully capture the complexity of the neurobiological basis behind disease progression[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTo address this limitation, we recently developed a structural MRI-based machine learning classifier[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e] that distinguished between faster and slower motor progressors in early PD with 89% accuracy, trained on a high-quality, well-structured longitudinal research dataset from the Parkinson\u0026rsquo;s Progression Markers Initiative (PPMI)[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. This model utilized a patient-specific multivariate gray matter volumetric distance (M\u003csub\u003eGMV\u003c/sub\u003e) combined with baseline demographic and clinical features to predict changes in Movement Disorder Society-Sponsored Revision of the Unified Parkinson's Disease Rating Scale Part III (MDS-UPDRS-III) scores over 48 months. Importantly, the model was trained and internally validated on a held-out test set from the PPMI dataset, using OFF-medication MDS-UPDRS-III assessments that reflect true disease severity without the confounding effects of medication.\u003c/p\u003e\u003cp\u003eFor such a model to be clinically meaningful, external validation in independent, real-world cohorts is essential. However, translating this approach to routine clinical datasets presents significant practical challenges. While OFF-medication motor assessments are standard in research protocols, patients are rarely evaluated in the OFF state in routine care due to time constraints, the discomfort of medication withdrawal, and the clinical focus on symptom management rather than standardized progression measurement[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Given these limitations, clinicians rely on practical indicators of disease progression, including patient-reported functional outcomes and medication dosage adjustments[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Patient-reported outcomes such as the MDS-UPDRS Part II (experiences of daily living) and the Schwab \u0026amp; England Activities of Daily Living (S\u0026amp;E ADL) scale are particularly valuable, as they capture functional decline from the patient\u0026rsquo;s perspective and are increasingly recognized as critical components of disease monitoring[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Moreover, as motor symptoms worsen, clinicians typically escalate medication doses. Consequently, longitudinal changes in medication requirements can serve as an indirect but practical indicator of motor progression[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. These changes are typically quantified using the Levodopa Equivalent Daily Dose (LEDD), a standardized metric that consolidates all antiparkinsonian medications into a single comparable value[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAlthough our prior work demonstrated MRI-based progression prediction using OFF-medication UPDRS-III scores, its applicability to routine clinical outcomes remains unclear. To address this gap, the present study examined whether patients stratified by our MRI-based classifier show differences in outcomes available within the PPMI cohort, specifically patient-reported functional measures and medication requirements. We hypothesized that individuals identified as \u0026lsquo;faster progressors\u0026rsquo; by the imaging model would experience greater loss of functional independence and require larger increases in medication dosage than those identified as \u0026lsquo;slower progressors.\u0026rsquo; In addition, we conducted secondary analyses to determine whether MRI-based classifications aligned with complementary indicators of disease worsening, including thresholds commonly regarded as clinically meaningful and a widely used global staging scale.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThis study utilized the same cohort as our previously published MRI-based machine learning classifier for predicting motor progression in PD. Comprehensive descriptions of the following methods have been reported previously[43], but key details are summarized here for clarity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eParticipants\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData were obtained from the PPMI database[23] (https://www.ppmi-info.org/access-data-specimens/download-data; RRID: SCR_006431) following standard application procedures. For current study details, please visit www.ppmi-info.org. The original study included 88 patients with early PD, selected according to inclusion and exclusion criteria described in our prior publication. Briefly, PD participants met the following inclusion criteria: recent diagnosis of PD with a positive DaTscan confirming the diagnosis; drug-na\u0026iuml;ve or no PD medication use at enrollment; and availability of baseline MRI data as well as MDS-UPDRS-III scores at baseline and at 48 months, with 48-month assessments performed in the OFF-medication state to minimize confounding from dopaminergic therapy. Exclusion criteria for all participants were a diagnosis of dementia or atypical parkinsonian syndromes, significant neurological or psychiatric conditions, and MRI artifacts or quality issues, including motion artifacts, field distortions, intensity inhomogeneities, or evidence of brain injury.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Parkinson\u0026rsquo;s Progression Markers Initiative (PPMI) study complies with the principles of the Declaration of Helsinki and Good Clinical Practice and is registered on ClinicalTrials.gov (NCT01141023). All participants provided written informed consent, and study protocols were approved by the local institutional review boards or ethics committees at each participating site (see list at https://www.ppmi-info.org/about-ppmi/ppmi-clinical-sites). One of the authors (AAV) obtained permission to use tier 3 PPMI data for this study, and all MRI as well as clinical data used for this study were downloaded in a de-identified format. The PPMI Data and Publications Committee reviewed our manuscript for administrative approval in accordance with PPMI policies. Therefore, the analyses presented in this article were conducted in accordance with approved PPMI guidelines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMRI Data Acquisition and Processing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eT1-weighted MRI images were acquired on 3.0T scanners from multiple vendors (Siemens, Philips, and GE) using standardized PPMI acquisition parameters (see MRI operations manual: ppmi-info.org/wp-content/uploads/2017/06/PPMI-MRI-Operations-Manual-V7.pdf). Image preprocessing was performed using the Computational Anatomy Toolbox 12 (version 12.8) within Statistical Parametric Mapping (SPM12), including bias field correction, tissue segmentation into gray matter, white matter, and cerebrospinal fluid, normalization to Montreal Neurological Institute space using the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra algorithm, and smoothing with an 8mm full width at half maximum Gaussian kernel[2]. Gray matter volumes (GMV) were extracted from 40 motor-specific regions of interest using the automated anatomical labeling atlas 3 (AAL3) and normalized by total intracranial volume[36]. Inter-scanner variability was addressed using NeuroHarmonize software with diagnosis, age, and sex as biological variables[32].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePatient-Specific Multivariate Gray Matter Volumetric Distance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe key innovation of our previous work was the development of a patient-specific summary score capturing GMV heterogeneity across multiple brain regions using Mahalanobis distance (M\u003csub\u003eGMV\u003c/sub\u003e)[43]. This multivariate approach calculated each patient\u0026apos;s distance from the healthy control distribution across all 40 motor-relevant brain regions simultaneously. Age was regressed as a covariate, and the normative reference was constructed from 120 age- and sex-matched healthy controls from PPMI with no history of neurological or psychiatric illness. M\u003csub\u003eGMV\u003c/sub\u003e was then computed as\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere s represents the vector of age-corrected GMV observations for a single patient, \u0026mu; is the vector of mean GMV values from healthy controls, and C is the covariance matrix between brain regions across healthy controls. To ensure robust estimation, M\u003csub\u003eGMV\u003c/sub\u003e was calculated relative to 1000 permutations of 100 randomly selected healthy controls, with median values reported.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine Learning Classification Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA support vector machine classifier with a radial basis function kernel was developed using M\u003csub\u003eGMV\u003c/sub\u003e, age, sex, and baseline MDS-UPDRS-III scores as input features. The model was trained to distinguish between \u0026quot;Slower Progressors\u0026quot; and \u0026quot;Faster Progressors\u0026quot; based on changes in OFF-medication MDS-UPDRS-III scores from baseline to 48 months. Patients were categorized using the median change score (\u0026Delta; = 9 points) as the threshold: slow progressors (n=46) had \u0026Delta;(MDS-UPDRS-III) \u0026le; 9, while fast progressors (n=42) had \u0026Delta;(MDS-UPDRS-III) \u0026gt; 9. The classifier achieved 89% accuracy with 87.5% sensitivity and 90.9% specificity on held-out test data (30% of the cohort), as detailed in our previous publication[43]. The Scikit-learn package of Python was used to develop the classifier[31].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIt is important to note that the present study did not involve retraining or modifying the machine learning model. The classifier parameters were fixed from our prior publication, and all analyses applied the pre-established model labels (\u0026ldquo;faster\u0026rdquo; vs \u0026ldquo;slower\u0026rdquo; progressors) to examine their relationship with new, independent clinical outcomes that were not used during model training. Therefore, the present work represents a secondary, hypothesis-driven internal validation of clinical translation rather than model redevelopment.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Outcome Measures\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo assess the clinical relevance of the MRI-based classification, we extracted additional progression markers from the same PPMI dataset, focusing on measures commonly available in routine practice. All measures were obtained at baseline and at the 48-month follow-up visit.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePrimary Outcome Measures\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eMDS-UPDRS Part II (Experiences of Daily Living):\u003c/strong\u003e This patient-reported outcome measures the motor-related impact of PD on daily activities across 13 items, scored from 0 (normal) to 52 (severe disability). Change scores from baseline to 48 months (\u0026Delta;MDS-UPDRS-II = MDS-UPDRS-II\u003csub\u003e48M\u003c/sub\u003e \u0026minus; MDS-UPDRS-II\u003csub\u003ebaseline\u003c/sub\u003e) were calculated to quantify functional decline from the patient\u0026rsquo;s perspective; positive values indicate decline[35].\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSchwab \u0026amp; England Activities of Daily Living Scale (\u003c/strong\u003e\u003cstrong\u003eS\u0026amp;E ADL)\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e The S\u0026amp;E ADL scale is a validated, widely recommended tool for assessing functional disability in PD. It rates overall functional capacity as a percentage of normal function (100% = completely independent; 0% = bedridden). Scores were recorded at baseline and at 48 months, with change calculated as \u0026Delta; S\u0026amp;E ADL = S\u0026amp;E ADL\u003csub\u003e48M\u003c/sub\u003e \u0026minus; S\u0026amp;E ADL\u003csub\u003ebaseline\u003c/sub\u003e. Negative values indicate functional decline[39].\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLevodopa Equivalent Daily Dose (LEDD):\u003c/strong\u003e The LEDD represents the cumulative dose of all PD medications converted to levodopa equivalents, providing a standardized measure of pharmacological treatment intensity[42]. As all patients were drug-na\u0026iuml;ve at baseline (LEDD = 0), LEDD values at the 48-month visit were used to assess cumulative medication burden, reflecting disease severity and treatment response over the study period.\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003eSecondary Outcome Measures\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn addition to the continuous measures above, we conducted analyses using clinically meaningful thresholds to evaluate whether the MRI-based classifier could predict categorical changes that are more directly interpretable in a clinical setting.\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eFor \u003cstrong\u003eMDS-UPDRS-II\u003c/strong\u003e, we used minimally clinically important difference thresholds from Horv\u0026aacute;th et al. (2017). A change of \u0026ge;2.51 points was classified as worsened, while \u0026lt;2.51 points indicated improved or stable[13].\u003c/li\u003e\n \u003cli\u003eFor \u003cstrong\u003eS\u0026amp;E ADL\u003c/strong\u003e, we applied a 10% decline threshold as a clinically significant cutoff, since one 10-point decrement represents the minimum clinically meaningful difference on this scale[6, 34, 40].\u003c/li\u003e\n \u003cli\u003eFor \u003cstrong\u003eLEDD\u003c/strong\u003e, we calculated an annualized slope (LEDD at 48 months \u0026divide; 4, expressed as mg/day/year; all participants were drug-na\u0026iuml;ve at baseline). We applied a threshold of \u0026ge;100 mg/day/year to define patients with substantially increased medication needs, based on prior reports[7, 44].\u003c/li\u003e\n \u003cli\u003eTo enhance clinical interpretability, we additionally included OFF-medication \u003cstrong\u003eHoehn \u0026amp; Yahr (HY)\u0026nbsp;\u003c/strong\u003estaging, a global clinician-rated marker of motor severity, to test whether machine learning classifications aligned with a universally recognized PD staging system. HY classifies PD severity into stages 1\u0026ndash;5, based on motor symptoms and functional disability. To capture meaningful worsening, we defined progression as an increase of \u0026ge;1 stage over 48 months (HY\u003csub\u003e48M\u003c/sub\u003e \u0026ndash; HY\u003csub\u003ebaseline\u003c/sub\u003e \u0026ge; 1). A one-stage shift is generally considered a relevant marker of advancing motor disability[9]. We also included an exploratory analysis to compare median HY stages at 48 months between progression groups.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBaseline characteristics were compared between slower and faster progressors using independent t-tests for continuous variables, chi-square tests for categorical variables, and Mann\u0026ndash;Whitney U tests for ordinal variables (HY stage). Primary analyses focused on three clinical outcomes at 48 months: LEDD change in MDS-UPDRS-II scores, and change in Schwab \u0026amp; England ADL scores. Effect sizes were calculated using Cohen\u0026rsquo;s d[8]. Secondary analyses evaluated whether our progression classification aligned with established definitions of clinically meaningful change by examining the proportion of patients meeting published thresholds or cutoffs, with group differences tested using chi-square. Odds ratios (ORs) with 95% confidence intervals were reported to quantify the strength of association between progression group and clinically meaningful outcomes. All analyses were conducted in Python (v3.12.11) using pandas, NumPy, SciPy, statsmodels, and matplotlib, with significance set at p \u0026lt; 0.05 after Benjamini\u0026ndash;Hochberg correction for multiple comparisons[4].\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eBaseline Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs shown in Table 1, there were no significant differences in age (60.43 \u0026plusmn; 9.81 vs. 59.75 \u0026plusmn; 9.05 years, \u003cem\u003ep\u003c/em\u003e = 0.66), sex distribution (78.6% vs. 58.7% male, \u003cem\u003ep\u003c/em\u003e = 0.06), MDS-UPDRS-II scores (4.69 \u0026plusmn; 3.13 vs. 5.74 \u0026plusmn; 4.57, \u003cem\u003ep\u003c/em\u003e = 0.16), S\u0026amp;E ADL scores (95.24% \u0026plusmn; 4.93 vs. 94.35% \u0026plusmn; 5.01, \u003cem\u003ep\u003c/em\u003e = 0.40), or HY stage (median [IQR]: 1 [1\u0026ndash;2] vs. 2 [1\u0026ndash;2], \u003cem\u003ep\u003c/em\u003e = 0.14). However, slower progressors demonstrated significantly higher MDS-UPDRS-III motor scores at baseline (22.41 \u0026plusmn; 9.04 vs. 16.71 \u0026plusmn; 7.55, \u003cem\u003ep\u003c/em\u003e = 0.002).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e Baseline Characteristics by MRI-Based Progression Classification\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"608\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFaster Progressors (n = 42)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSlower Progressors (n = 46)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003eAge, years (mean \u0026plusmn; SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e60.43 \u0026plusmn; 9.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e59.75 \u0026plusmn; 9.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003eSex, male n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e33 (78.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e27 (58.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003eMDS-UPDRS-II baseline\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e4.69 \u0026plusmn; 3.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e5.74 \u0026plusmn; 4.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.16\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003eMDS-UPDRS-III baseline\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e16.71 \u0026plusmn; 7.55\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e22.41 \u0026plusmn; 9.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003eS\u0026amp;E ADL baseline, %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e95.24 \u0026plusmn; 4.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e94.35 \u0026plusmn; 5.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 181px;\"\u003e\n \u003cp\u003eHY baseline, median (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e1 (1\u0026ndash;2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e2 (1\u0026ndash;2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eValues are mean \u0026plusmn; standard deviation (SD) except for HY, where a non-parametric test was employed. n (%), or median (interquartile range, IQR) as appropriate.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAbbreviations: MDS-UPDRS-II = Movement Disorder Society\u0026ndash;sponsored revision of the Unified Parkinson\u0026rsquo;s Disease Rating Scale Part II; MDS-UPDRS-III = Movement Disorder Society\u0026ndash;sponsored revision of the Unified Parkinson\u0026rsquo;s Disease Rating Scale Part III; S\u0026amp;E ADL = Schwab \u0026amp; England Activities of Daily Living; HY = Hoehn \u0026amp; Yahr.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Progression at 48 Months\u0026mdash;Primary Analyses\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAt 48 months, faster progressors demonstrated significantly greater clinical deterioration across all primary outcome measures compared to slower progressors (Table 2). The mean increase in MDS-UPDRS-II scores was 5.31 \u0026plusmn; 4.77 for faster progressors versus 2.76 \u0026plusmn; 4.56 for slower progressors (\u003cem\u003ep\u003c/em\u003e = 0.01, Cohen\u0026apos;s d = 0.54). S\u0026amp;E ADL independence declined by 9.29 \u0026plusmn; 8.12% in faster progressors compared to 4.89 \u0026plusmn; 6.18% in slower progressors (\u003cem\u003ep\u003c/em\u003e = 0.009, Cohen\u0026apos;s d = 0.57). LEDD requirements at 48 months were 423.45 \u0026plusmn; 274.16 mg/day for faster progressors versus 278.37 \u0026plusmn; 203.84 mg/day for slower progressors (\u003cem\u003ep\u003c/em\u003e = 0.006, Cohen\u0026apos;s d = 0.60). All effect sizes indicated moderate to large clinically meaningful differences between groups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e Clinical Progression Markers at 48 Months by MRI-Based Classification\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutcome Measure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eFaster Progressors (n=42)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSlower Progressors (n=46)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCohen\u0026apos;s d\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026Delta; MDS-UPDRS-II\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.31 \u0026plusmn; 4.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.76 \u0026plusmn; 4.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.01\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026Delta; S\u0026amp;E ADL, %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026minus;9.29 \u0026plusmn; 8.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026minus;4.89 \u0026plusmn; 6.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.009\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEDD at 48M, mg/day\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e423.45 \u0026plusmn; 274.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e278.37 \u0026plusmn; 203.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.006\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eData presented as mean \u0026plusmn; standard deviation. \u0026Delta; indicates change from baseline to 48 months.\u003c/p\u003e\n\u003cp\u003eAbbreviations: MDS-UPDRS-II = Movement Disorder Society\u0026ndash;sponsored revision of the Unified Parkinson\u0026rsquo;s Disease Rating Scale, Part II (Motor Aspects of Experiences of Daily Living); S\u0026amp;E ADL = Schwab \u0026amp; England Activities of Daily Living. LEDD = Levodopa Equivalent Daily Dose.\u003c/p\u003e\n\u003cp\u003eStatistical significance at \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Meaningfulness Analysis\u0026mdash;Secondary Analyses\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSecondary analyses revealed distinct patterns of clinically meaningful progression between groups (Table 3; Figure 1). Faster progressors were significantly more likely to experience worsening in daily functioning, with 81% exceeding the threshold for clinically meaningful decline on the MDS-UPDRS-II compared to 52% of slower progressors (OR = 3.90, 95% CI 1.49\u0026ndash;10.21, \u003cem\u003ep\u003c/em\u003e = 0.009). A trend was observed for activities of daily living, where 62% of faster progressors showed \u0026ge;10% decline on the S\u0026amp;E ADL scale versus 41% of slower progressors (OR = 2.31, 95% CI 0.98\u0026ndash;5.43, \u003cem\u003ep\u003c/em\u003e = 0.05). With respect to medication use, 48% of faster progressors exceeded the annualized LEDD slope threshold of \u0026ge;100 mg/day/year compared to 30% of slower progressors (OR = 2.08, 95% CI 0.87\u0026ndash;4.97, \u003cem\u003ep\u003c/em\u003e = 0.09), which was not statistically significant. The most pronounced difference was observed in HY staging, where 62% of faster progressors experienced \u0026ge;1 stage worsening over 48 months, compared to only 28% of slower progressors (OR = 4.13, 95% CI 1.69\u0026ndash;10.09, \u003cem\u003ep\u003c/em\u003e = 0.003; Figure 2).\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eSupporting this finding, exploratory analysis revealed that faster progressors had higher HY stages at 48 months (median [IQR] 2 [2\u0026ndash;2]) compared to slower progressors (2 [1\u0026ndash;2], \u003cem\u003ep\u003c/em\u003e = 0.012). The distributional shifts in HY stages between faster and slower progressors are provided in Supplementary File 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e Clinically Meaningful Progression Thresholds at 48 Months by MRI-Based Classification.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"680\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutcome Measure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eThresholds\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eFaster Progressors (n=42)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSlower Progressors (n=46)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eOdds ratio\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMDS-UPDRS-II\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026Delta; \u0026lt; 2.51\u003c/p\u003e\n \u003cp\u003e\u0026Delta; \u0026ge; 2.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (19%)\u003c/p\u003e\n \u003cp\u003e34 (81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22 (48%)\u003c/p\u003e\n \u003cp\u003e24 (52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e3.90 (1.49\u0026ndash;10.21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e0.009\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eS\u0026amp;E ADL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026Delta; \u0026lt; 10%\u003c/p\u003e\n \u003cp\u003e\u0026Delta; \u0026ge; 10%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (38%)\u003c/p\u003e\n \u003cp\u003e26 (62%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e27 (59%)\u003c/p\u003e\n \u003cp\u003e19 (41%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2.31 (0.98\u0026ndash;5.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cem\u003e0.05\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLEDD slope\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; 100 mg/day/year\u003c/p\u003e\n \u003cp\u003e\u0026ge; 100 mg/day/year\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22 (52%)\u003c/p\u003e\n \u003cp\u003e20 (48%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e32 (70%)\u003c/p\u003e\n \u003cp\u003e14 (30%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2.08 (0.87\u0026ndash;4.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0.09\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHY\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026Delta; \u0026lt; 1\u003c/p\u003e\n \u003cp\u003e\u0026Delta; \u0026ge; 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (38%)\u003c/p\u003e\n \u003cp\u003e26 (62%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e33 (72%)\u003c/p\u003e\n \u003cp\u003e13 (28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e4.13 (1.69\u0026ndash;10.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eValues are n (%). \u0026Delta; = change from baseline to 48 months.\u003c/p\u003e\n\u003cp\u003eAbbreviations: CI = Confidence Interval; MDS-UPDRS-II = Movement Disorder Society\u0026ndash;sponsored revision of the Unified Parkinson\u0026rsquo;s Disease Rating Scale Part II; S\u0026amp;E ADL = Schwab \u0026amp; England Activities of Daily Living; LEDD = Levodopa Equivalent Daily Dose; HY = Hoehn \u0026amp; Yahr.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStatistical significance at \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we evaluated whether our previously developed machine learning classifier[43] could predict pragmatic outcomes commonly available in clinical practice. The classifier was originally trained on 48-month OFF-medication MDS-UPDRS Part III change scores to stratify patients as faster or slower motor progressors. Here, we performed an internal validation within the PPMI cohort, applying the classifier to new, clinically relevant outcomes not used in model training. Patients identified as faster progressors exhibited greater functional decline and pharmacologic burden compared to slower progressors, providing preliminary support for the potential utility of structural MRI-based prognostic models for predicting disease progression in early PD.An intriguing finding emerged when comparing baseline characteristics: slower progressors had significantly higher baseline MDS-UPDRS-III motor scores compared to faster progressors (22.41 vs. 16.71, p = 0.002). While this initially appears counterintuitive, it may reflect important underlying pathophysiological mechanisms. This paradox suggests that baseline motor severity does not necessarily reflect an individual patient’s progression trajectory, and that underlying neurobiological mechanisms captured by MRI may be more predictive of future decline. These findings point toward the existence of distinct PD subgroups, in which structural brain alterations differentially shape the relationship between clinical presentation and subsequent disease trajectory.\u003c/p\u003e\n\u003cp\u003eTo further evaluate the clinical relevance of these subgroups, we assessed outcomes that directly reflect patient function and independence. MDS-UPDRS-II and S\u0026amp;E ADL represent medication-independent assessments of functional decline, making them valuable tools for monitoring progression[11, 37, 39]. We observed that faster progressors demonstrated significantly greater deterioration in ΔMDS-UPDRS-II scores, with 81% experiencing a clinically meaningful decline compared with 52% of slower progressors. This finding was consistent with ΔS\u0026amp;E ADL scores, where faster progressors showed greater loss of independence (62% with ≥10% decline vs 41% of slower progressors). Although this difference reached only borderline significance (\u003cem\u003ep\u003c/em\u003e = 0.05), it reflected the broader pattern of functional deterioration across complementary domains. Unlike motor examinations (MDS-UPDRS-III), which require trained raters and are medication status-dependent, both MDS-UPDRS Part II and S\u0026amp;E ADL can be completed directly by patients, supporting their utility for routine monitoring[29]. The ability of our imaging classifier to capture decline across multiple patient-centered domains underscores its translational potential in clinical practice.\u003c/p\u003e\n\u003cp\u003eIn addition to functional deterioration, those classified as faster progressors required increased medication over time. At 48 months, they had significantly higher LEDD than slower progressors, consistent with prior evidence that worsening motor symptoms drive increased treatment needs[3, 21]. Supporting this pattern, Birkenbihl et al. also reported that fast-progressing patients required higher LEDD after approximately four years of follow-up[5]. Together, these findings support the notion that patients at higher risk for rapid motor progression demand greater treatment intensity over time. Importantly, greater medication use did not mitigate decline, but rather co-occurred with it, underscoring the robustness of the classifier’s predictions.\u003c/p\u003e\n\u003cp\u003eDespite these promising results, our clinically meaningful threshold-based analysis revealed important limitations. Using a cutoff derived from prior literature (~100 mg/day/year)[7, 44], the non-significant trend in LEDD progression (\u003cem\u003ep\u0026nbsp;\u003c/em\u003e= 0.09) warrants careful interpretation. While LEDD is widely considered a surrogate for disease severity[42], it is strongly influenced by prescribing practices, including the timing of levodopa initiation, choice of adjunct therapies, and responsiveness to patient complaints[18, 19, 33]. Moreover, conversion formulas are approximations that introduce variability,[16, 26] with scaling differences depending on the weight assigned to each drug class. As a result, LEDD progression may reflect both disease severity and physician prescribing patterns, which complicates its interpretation as a marker of clinical progression[27, 41]. This inherent confounding suggests caution in using medication burden as a primary endpoint for evaluating progression prediction models. Future studies should prioritize outcomes that are more directly tied to disease biology and less dependent on prescriber variability.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo assess our classifier's discriminative ability across different clinical measures, we explored whether it could identify differences in HY staging. While HY is also medication-dependent like motor examinations (MDS-UPDRS-III), it represents a universally recognized staging system widely used in clinical practice, making it a valuable benchmark for our model. Notably, HY staging revealed the most pronounced group difference, with 62% of faster progressors experiencing a ≥1-stage worsening compared with only 28% of slower progressors. Although the scale is coarse and ordinal, a one-stage increase is generally regarded as a clinically meaningful indicator of symptom progression in PD[9, 28]. The ability of our model to stratify patients by their risk of HY progression highlights a key translational strength by connecting imaging-based predictions to a simple clinical tool already central to practice. We also compared HY scores at 48 months, which confirmed higher staging among faster progressors. This reinforces that HY progression differences were robust and not solely dependent on categorical thresholds.\u003c/p\u003e\n\u003cp\u003eConventionally, structural T1-weighted MRI is not considered useful for prognosis in PD, as routine scans often appear “normal”[22]. In our original work, however, we introduced a patient-specific multivariate gray matter volumetric distance (M\u003csub\u003eGMV\u003c/sub\u003e) derived using Mahalanobis distance, which summarized atrophy in the gray matter regions across multiple regions into a single score[43]. When combined with machine learning, this approach demonstrated that even standard MRI data can extract prognostically valuable information by identifying patients at risk for faster motor deterioration. Importantly, while that model was trained and internally validated using OFF-medication MDS-UPDRS-III scores, direct external validation remains limited because OFF assessments are rarely available in clinical practice. The present findings suggest that future prognostic models need not rely exclusively on such OFF assessments. Instead, pragmatic measures such as MDS-UPDRS-II and the S\u0026amp;E ADL scale—more commonly collected in routine care—may serve as equally meaningful and clinically relevant endpoints. This shift would allow machine learning models to undergo both rigorous internal and external validation in real-world datasets[12]. Taken together, our work underscores that baseline T1-weighted MRI, often deemed clinically uninformative, can provide a powerful substrate for prognostic modeling when coupled with advanced analytic methods.\u003c/p\u003e\n\u003cp\u003eAlthough these findings show promise, future development must take into account several limitations of the current study. First, while this work was conducted in the well-characterized PPMI research cohort[23], future studies should evaluate such models in clinical datasets, where real-world challenges such as variability in imaging protocols, treatment strategies, and follow-up are unavoidable. In particular, external validation in non-PPMI cohorts, including hospital-based populations with more heterogeneous data, will be critical to establish generalizability. Second, our original classifier was trained on only 88 patients, reflecting the relatively small subset of PPMI participants with complete OFF-medication MDS-UPDRS-III data. In contrast, other clinical measures such as MDS-UPDRS-II or S\u0026amp;E ADL are more widely available both in PPMI and in routine practice, and anchoring models to these outcomes would enable training on larger and more representative samples. Finally, the 48-month follow-up period may not capture the full course of disease progression. Extended follow-up will be critical to establish whether predictive performance persists and continues to hold clinical relevance over time. Addressing these limitations will be essential for developing MRI-based prognostic tools that are both biologically grounded and clinically applicable in everyday PD care.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGiven the wide spectrum of PD progression ranging from relatively benign to rapidly advancing forms, the ability to anticipate disease course carries substantial clinical and research value. Early identification of likely faster progressors could assist clinicians in tailoring follow-up intensity, optimizing therapeutic timing, and counselling patients and caregivers. Prognostic stratification also holds implications for clinical trial design—facilitating enrichment of disease-modifying trials with higher-risk participants—and may ultimately support the discovery and validation of biomarkers linked to progression. Therefore, a reliable and accessible predictor of disease course addresses a critical unmet need in PD care and research.\u003c/p\u003e\n\u003cp\u003eIn conclusion, this study demonstrated that an MRI-based machine learning classifier, originally trained to predict OFF-medication motor progression, successfully identified patients at risk for clinically meaningful deterioration across multiple real-world outcome measures. Patients classified as faster progressors exhibited significantly greater functional decline on patient-reported measures, higher medication requirements, and more pronounced staging progression compared to slower progressors. Together, these findings represent an important step toward establishing the clinical utility of structural MRI–based prognostic models for predicting disease progression in early PD. Such prognostic information could inform clinical trial stratification, guide patient counseling, and potentially support decisions about treatment timing.\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData used in the preparation of this article were obtained on 2025-07-28 from the Parkinson’s Progression Markers Initiative (PPMI) database (https://www.ppmi-info.org/access-data-specimens/download-data), RRID:SCR_006431. For up-to-date information on the study, visit http://www.ppmi-info.org.\u003c/p\u003e\n\u003cp\u003ePPMI – a public-private partnership – is funded by the Michael J. Fox Foundation for Parkinson’s Research and funding partners, including 4D Pharma, Abbvie, AcureX, Allergan, Amathus Therapeutics, Aligning Science Across Parkinson's, AskBio, Avid Radiopharmaceuticals, BIAL, BioArctic, Biogen, Biohaven, BioLegend, BlueRock Therapeutics, Bristol-Myers Squibb, Calico Labs, Capsida Biotherapeutics, Celgene, Cerevel Therapeutics, Coave Therapeutics, DaCapo Brainscience, Denali, Edmond J. Safra Foundation, Eli Lilly, Gain Therapeutics, GE HealthCare, Genentech, GSK, Golub Capital, Handl Therapeutics, Insitro, Jazz Pharmaceuticals, Johnson \u0026amp; Johnson Innovative Medicine, Lundbeck, Merck, Meso Scale Discovery, Mission Therapeutics, Neurocrine Biosciences, Neuron23, Neuropore, Pfizer, Piramal, Prevail Therapeutics, Roche, Sanofi, Servier, Sun Pharma Advanced Research Company, Takeda, Teva, UCB, Vanqua Bio, Verily, Voyager Therapeutics, the Weston Family Foundation and Yumanity Therapeutics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA.A. Vijayakumari\u003c/strong\u003e: study concept and design; data extraction; data analysis and interpretation; drafting and revision of the manuscript for content.\u003cbr\u003e\u003cstrong\u003eD. Teixeira-Dos-Santos\u003c/strong\u003e: critical revision of the manuscript for content; input on study interpretation.\u003cbr\u003e\u003cstrong\u003eH.H. Fernandez\u003c/strong\u003e: study concept and design; critical revision of the manuscript for content.\u003cbr\u003e\u003cstrong\u003eB.L. Walter\u003c/strong\u003e: study concept and design; critical revision of the manuscript for content.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data analyzed in this study are available from the Parkinson’s Progression Markers Initiative (PPMI) database (https://www.ppmi-info.org/access-data-specimens/download-data; RRID:SCR_006431) upon application and approval by the PPMI Data and Publications Committee. They are also available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no competing interests to declare that are relevant to the contents of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by a Cleveland Clinic Neurological Institute Clinician Scientist Career Award (A.A.V.).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eArmstrong MJ, Okun MS (2020) Diagnosis and Treatment of Parkinson Disease: A Review. Jama 323:548-560\u003c/li\u003e\n\u003cli\u003eAshburner J, Friston KJ (2005) Unified segmentation. Neuroimage 26:839-851\u003c/li\u003e\n\u003cli\u003eBartl M, Dakna M, Schade S, Wicke T, Lang E, Ebentheuer J, Weber S, Trenkwalder C, Mollenhauer B (2021) Longitudinal Change and Progression Indicators Using the Movement Disorder Society-Unified Parkinson\u0026rsquo;s Disease Rating Scale in Two Independent Cohorts with Early Parkinson\u0026rsquo;s Disease. Journal of Parkinson\u0026rsquo;s Disease 12:437-452\u003c/li\u003e\n\u003cli\u003eBenjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B (Methodological) 57:289-300\u003c/li\u003e\n\u003cli\u003eBirkenbihl C, Ahmad A, Massat NJ, Raschka T, Avbersek A, Downey P, Armstrong M, Fr\u0026ouml;hlich H (2023) Artificial intelligence-based clustering and characterization of Parkinson\u0026apos;s disease trajectories. Scientific Reports 13:2897\u003c/li\u003e\n\u003cli\u003eBou\u0026ccedil;a-Machado R, Fernandes A, Ranzato C, Beneby D, Nzwalo H, Ferreira JJ (2022) Measurement tools to assess activities of daily living in patients with Parkinson\u0026apos;s disease: A systematic review. Front Neurosci 16:945398\u003c/li\u003e\n\u003cli\u003eBovenzi R, Conti M, Degoli GR, Cerroni R, Simonetta C, Liguori C, Salimei C, Pisani A, Pierantozzi M, Stefani A, Mercuri NB, Schirinzi T (2023) Shaping the course of early-onset Parkinson\u0026apos;s disease: insights from a longitudinal cohort. Neurol Sci 44:3151-3159\u003c/li\u003e\n\u003cli\u003eCohen J (2013) Statistical power analysis for the behavioral sciences. routledge\u003c/li\u003e\n\u003cli\u003eGoetz CG, Poewe W, Rascol O, Sampaio C, Stebbins GT, Counsell C, Giladi N, Holloway RG, Moore CG, Wenning GK, Yahr MD, Seidl L (2004) Movement Disorder Society Task Force report on the Hoehn and Yahr staging scale: status and recommendations. Mov Disord 19:1020-1028\u003c/li\u003e\n\u003cli\u003eGreenland JC, Williams-Gray CH, Barker RA (2019) The clinical heterogeneity of Parkinson\u0026apos;s disease and its therapeutic implications. Eur J Neurosci 49:328-338\u003c/li\u003e\n\u003cli\u003eHarrison MB, Wylie SA, Frysinger RC, Patrie JT, Huss DS, Currie LJ, Wooten GF (2009) UPDRS activity of daily living score as a marker of Parkinson\u0026apos;s disease progression. Mov Disord 24:224-230\u003c/li\u003e\n\u003cli\u003eHo SY, Phua K, Wong L, Bin Goh WW (2020) Extensions of the External Validation for Checking Learned Model Interpretability and Generalizability. Patterns (N Y) 1:100129\u003c/li\u003e\n\u003cli\u003eHorv\u0026aacute;th K, Aschermann Z, Kov\u0026aacute;cs M, Makkos A, Harmat M, Janszky J, Komoly S, Kar\u0026aacute;di K, Kov\u0026aacute;cs N (2017) Minimal clinically important differences for the experiences of daily living parts of movement disorder society-sponsored unified Parkinson\u0026apos;s disease rating scale. Mov Disord 32:789-793\u003c/li\u003e\n\u003cli\u003eHoskovcov\u0026aacute; M, Du\u0026scaron;ek P, Sieger T, Brožov\u0026aacute; H, Z\u0026aacute;rubov\u0026aacute; K, Bezd\u0026iacute;ček O, \u0026Scaron;prdl\u0026iacute;k O, Jech R, \u0026Scaron;tochl J, Roth J, Růžička E (2015) Predicting Falls in Parkinson Disease: What Is the Value of Instrumented Testing in OFF Medication State? PLoS One 10:e0139849\u003c/li\u003e\n\u003cli\u003eJankovic J, McDermott M, Carter J, Gauthier S, Goetz C, Golbe L, Huber S, Koller W, Olanow C, Shoulson I, et al. (1990) Variable expression of Parkinson\u0026apos;s disease: a base-line analysis of the DATATOP cohort. The Parkinson Study Group. Neurology 40:1529-1534\u003c/li\u003e\n\u003cli\u003eJost ST, Kaldenbach MA, Antonini A, Martinez-Martin P, Timmermann L, Odin P, Katzenschlager R, Borgohain R, Fasano A, Stocchi F, Hattori N, Kukkle PL, Rodr\u0026iacute;guez-Violante M, Falup-Pecurariu C, Schade S, Petry-Schmelzer JN, Metta V, Weintraub D, Deuschl G, Espay AJ, Tan EK, Bhidayasiri R, Fung VSC, Cardoso F, Trenkwalder C, Jenner P, Ray Chaudhuri K, Dafsari HS (2023) Levodopa Dose Equivalency in Parkinson\u0026apos;s Disease: Updated Systematic Review and Proposals. Mov Disord 38:1236-1252\u003c/li\u003e\n\u003cli\u003eKalia LV, Lang AE (2015) Parkinson\u0026apos;s disease. Lancet 386:896-912\u003c/li\u003e\n\u003cli\u003eKalilani L, Friesen D, Boudiaf N, Asgharnejad M (2019) The characteristics and treatment patterns of patients with Parkinson\u0026apos;s disease in the United States and United Kingdom: A retrospective cohort study. PLoS One 14:e0225723\u003c/li\u003e\n\u003cli\u003eKu M, Je NK (2023) Exploring the prescribing trends and factors affecting initial anti-parkinsonian drug selection in Korea: A nationwide population-based cohort study. Journal of Clinical Neuroscience 116:60-66\u003c/li\u003e\n\u003cli\u003eLim S-Y, Tan AH (2018) Historical perspective: The pros and cons of conventional outcome measures in Parkinson\u0026apos;s disease. Parkinsonism \u0026amp; Related Disorders 46:S47-S52\u003c/li\u003e\n\u003cli\u003eLin YH, Fang TC, Lei HB, Chiu SC, Chang MH, Guo YJ (2023) UPSIT subitems may predict motor progression in Parkinson\u0026apos;s disease. Front Neurol 14:1265549\u003c/li\u003e\n\u003cli\u003eMahlknecht P, Hotter A, Hussl A, Esterhammer R, Schocke M, Seppi K (2010) Significance of MRI in diagnosis and differential diagnosis of Parkinson\u0026apos;s disease. Neurodegener Dis 7:300-318\u003c/li\u003e\n\u003cli\u003eMarek K, Chowdhury S, Siderowf A, Lasch S, Coffey CS, Caspell-Garcia C, Simuni T, Jennings D, Tanner CM, Trojanowski JQ, Shaw LM, Seibyl J, Schuff N, Singleton A, Kieburtz K, Toga AW, Mollenhauer B, Galasko D, Chahine LM, Weintraub D, Foroud T, Tosun-Turgut D, Poston K, Arnedo V, Frasier M, Sherer T (2018) The Parkinson\u0026apos;s progression markers initiative (PPMI) - establishing a PD biomarker cohort. Ann Clin Transl Neurol 5:1460-1477\u003c/li\u003e\n\u003cli\u003eMartin H, Ali R, Sriram A, Coleman R, Ruether E, Boyce H, Maley ML, Khan M (2025) Leukoaraiosis does not impact motor outcomes in Parkinson\u0026apos;s patients post deep brain stimulation. Clin Park Relat Disord 12:100348\u003c/li\u003e\n\u003cli\u003eMuslimovic D, Post B, Speelman JD, Schmand B, de Haan RJ (2008) Determinants of disability and quality of life in mild to moderate Parkinson disease. Neurology 70:2241-2247\u003c/li\u003e\n\u003cli\u003eNyholm D, Jost WH (2021) An updated calculator for determining levodopa-equivalent dose. Neurological Research and Practice 3:58\u003c/li\u003e\n\u003cli\u003eOrayj K, Lane E (2019) Patterns and Determinants of Prescribing for Parkinson\u0026apos;s Disease: A Systematic Literature Review. Parkinsons Dis 2019:9237181\u003c/li\u003e\n\u003cli\u003ePagano G, De Micco R, Yousaf T, Wilson H, Chandra A, Politis M (2018) REM behavior disorder predicts motor progression and cognitive decline in Parkinson disease. Neurology 91:e894-e905\u003c/li\u003e\n\u003cli\u003eParashos SA, Luo S, Biglan KM, Bodis-Wollner I, He B, Liang GS, Ross GW, Tilley BC, Shulman LM (2014) Measuring disease progression in early Parkinson disease: the National Institutes of Health Exploratory Trials in Parkinson Disease (NET-PD) experience. JAMA Neurol 71:710-716\u003c/li\u003e\n\u003cli\u003ePassaretti M, Ver\u0026eacute;b D, Mijalkov M, Chang Y-W, Zhao H, Zufiria-Gerbol\u0026eacute;s B, Sun J, Volpe G, Rivera N, Bologna M, Pereira JB (2025) Clinical progression and genetic pathways in body-first and brain-first Parkinson\u0026rsquo;s disease. Molecular Neurodegeneration 20:74\u003c/li\u003e\n\u003cli\u003ePedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss R, Dubourg V, Vanderplas J, Passos A, Cournapeau D, Brucher M, Perrot M, Duchesnay \u0026Eacute; (2011) Scikit-learn: Machine learning in Python. Journal of machine Learning research 12: 2825-2830\u003c/li\u003e\n\u003cli\u003ePomponio R, Erus G, Habes M, Doshi J, Srinivasan D, Mamourian E, Bashyam V, Nasrallah IM, Satterthwaite TD, Fan Y, Launer LJ, Masters CL, Maruff P, Zhuo C, Volzke H, Johnson SC, Fripp J, Koutsouleris N, Wolf DH, Gur R, Gur R, Morris J, Albert MS, Grabe HJ, Resnick SM, Bryan RN, Wolk DA, Shinohara RT, Shou H, Davatzikos C (2020) Harmonization of large MRI datasets for the analysis of brain imaging patterns throughout the lifespan. Neuroimage 208:116450\u003c/li\u003e\n\u003cli\u003ePringsheim T, Day GS, Smith DB, Rae-Grant A, Licking N, Armstrong MJ, de Bie RMA, Roze E, Miyasaki JM, Hauser RA, Espay AJ, Martello JP, Gurwell JA, Billinghurst L, Sullivan K, Fitts MS, Cothros N, Hall DA, Rafferty M, Hagerbrant L, Hastings T, O\u0026apos;Brien MD, Silsbee H, Gronseth G, Lang AE (2021) Dopaminergic Therapy for Motor Symptoms in Early Parkinson Disease Practice Guideline Summary: A Report of the AAN Guideline Subcommittee. Neurology 97:942-957\u003c/li\u003e\n\u003cli\u003eRamaker C, Marinus J, Stiggelbout AM, van Hilten BJ (2002) Systematic evaluation of rating scales for impairment and disability in Parkinson\u0026rsquo;s disease. Movement Disorders 17:867-876\u003c/li\u003e\n\u003cli\u003eRodriguez-Blazquez C, Rojo-Abuin JM, Alvarez-Sanchez M, Arakaki T, Bergareche-Yarza A, Chade A, Garretto N, Gershanik O, Kurtis MM, Martinez-Castrillo JC, Mendoza-Rodriguez A, Moore HP, Rodriguez-Violante M, Singer C, Tilley BC, Huang J, Stebbins GT, Goetz CG, Martinez-Martin P (2013) The MDS-UPDRS Part II (motor experiences of daily living) resulted useful for assessment of disability in Parkinson\u0026apos;s disease. Parkinsonism Relat Disord 19:889-893\u003c/li\u003e\n\u003cli\u003eRolls ET, Huang CC, Lin CP, Feng J, Joliot M (2020) Automated anatomical labelling atlas 3. Neuroimage 206:116189\u003c/li\u003e\n\u003cli\u003eSampaio C (2009) Can focusing on UPDRS Part II make assessments of Parkinson disease progression more efficient? Nature Reviews Neurology 5:130-131\u003c/li\u003e\n\u003cli\u003eSchrag A, Sampaio C, Counsell N, Poewe W (2006) Minimal clinically important change on the unified Parkinson\u0026apos;s disease rating scale. Mov Disord 21:1200-1207\u003c/li\u003e\n\u003cli\u003eSchwab RS, England AC, Jr. (1969) Projection technique for evaluating surgery in Parkinson\u0026apos;s disease. In: Gillingham FJ, Donaldson IML (eds) Third Symposium on Parkinson\u0026rsquo;s Disease. E \u0026amp; S Livingstone, Edinburgh, Scotland, pp 152-157\u003c/li\u003e\n\u003cli\u003eSiderowf A (2010) Schwab and England Activities of Daily Living Scale. In: Kompoliti K, Verhagen L (eds) Encyclopedia of Movement Disorders. Elsevier, Amsterdam, pp 99-100\u003c/li\u003e\n\u003cli\u003eTolosa E, Ebersbach G, Ferreira JJ, Rascol O, Antonini A, Foltynie T, Gibson R, Magalhaes D, Rocha JF, Lees A (2021) The Parkinson\u0026apos;s Real-World Impact Assessment (PRISM) Study: A European Survey of the Burden of Parkinson\u0026apos;s Disease in Patients and their Carers. J Parkinsons Dis 11:1309-1323\u003c/li\u003e\n\u003cli\u003eTomlinson CL, Stowe R, Patel S, Rick C, Gray R, Clarke CE (2010) Systematic review of levodopa dose equivalency reporting in Parkinson\u0026apos;s disease. Mov Disord 25:2649-2653\u003c/li\u003e\n\u003cli\u003eVijayakumari AA, Fernandez HH, Walter BL (2023) MRI-based multivariate gray matter volumetric distance for predicting motor symptom progression in Parkinson\u0026apos;s disease. Scientific Reports 13:17704\u003c/li\u003e\n\u003cli\u003eWilson J, Alcock L, Yarnall AJ, Lord S, Lawson RA, Morris R, Taylor JP, Burn DJ, Rochester L, Galna B (2020) Gait Progression Over 6 Years in Parkinson\u0026apos;s Disease: Effects of Age, Medication, and Pathology. Front Aging Neurosci 12:577435\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Cleveland Clinic","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Parkinson’s disease, Artificial Intelligence, Machine learning, MRI, Parkinson’s prediction, Neuroimaging","lastPublishedDoi":"10.21203/rs.3.rs-7870464/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7870464/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e Parkinson’s disease (PD) shows marked variability in disease progression, and predicting individual trajectories remains challenging. We previously developed a structural MRI–based machine learning classifier that distinguished faster from slower motor progressors using OFF-medication Movement Disorder Society–Unified Parkinson’s Disease Rating Scale Part III (MDS-UPDRS-III) scores with 89% accuracy. As OFF assessments are rarely performed in clinical practice, we evaluated whether this classifier predicts outcomes routinely used in care.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e Eighty-eight early PD patients from the Parkinson’s Progression Markers Initiative were previously classified as faster (n=42) or slower (n=46) motor progressors using a support vector machine model incorporating patient-specific multivariate gray matter volumetric distance and baseline clinical features. Primary outcomes were 48-month changes (Δ) in MDS-UPDRS Part II (experiences of daily living), Schwab \u0026amp; England Activities of Daily Living (S\u0026amp;E ADL), and levodopa equivalent daily dose (LEDD). Secondary analyses examined clinically meaningful thresholds: MDS-UPDRS-II worsening (≥2.51 points), ≥10% S\u0026amp;E ADL decline, ≥100 mg/day/year LEDD slope, and ≥1-stage Hoehn \u0026amp; Yahr (HY) progression.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e Faster progressors showed significantly greater functional decline (ΔMDS-UPDRS-II: 5.31±4.77 vs. 2.76±4.56, p=0.01; ΔS\u0026amp;E ADL: 9.29±8.12% vs. 4.89±6.18%, p=0.009) and higher medication requirements (LEDD: 423.45±274.16 vs. 278.37±203.84 mg/day, p=0.006). Clinically meaningful deterioration was more frequent among faster progressors for MDS-UPDRS-II (81% vs. 52%, OR=3.90, p=0.009) and HY staging (62% vs. 28%, OR=4.13, p=0.003).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e An MRI-based classifier trained on OFF-medication motor assessments successfully predicts clinically meaningful deterioration across multiple real-world outcomes, supporting its potential utility for prognostic stratification in early PD.\u003c/p\u003e","manuscriptTitle":"Evaluating an MRI-Based Machine Learning Classifier for Parkinson’s Progression Using Real-World Clinical Measures","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-17 12:46:43","doi":"10.21203/rs.3.rs-7870464/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c116c950-29cd-403b-9668-241861d2737c","owner":[],"postedDate":"October 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":56362662,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-10-17T12:46:43+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-17 12:46:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7870464","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7870464","identity":"rs-7870464","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}