This retrospective study was conducted by reviewing medical records of patients evaluated for parkinsonism at The First Hospital of Hebei Medical University between December 2022 and January 2025. The final study cohort for statistical modeling consisted of 173 patients: 102 diagnosed with PD and 71 with PS. These diagnoses were established by experienced neurologists specializing in movement disorders, based on comprehensive clinical evaluation including motor and non-motor symptoms, response to levodopa where applicable, and longitudinal follow-up for at least 2 years to confirm diagnostic stability. Specifically, the diagnosis of PD was made according to the 2015 Movement Disorder Society (MDS) clinical diagnostic criteria for Parkinson’s disease (2), while diagnoses for PS subtypes were based on their respective international consensus criteria; specifically, Multiple System Atrophy (MSA) was diagnosed according to the Movement Disorder Society criteria (14), Progressive Supranuclear Palsy (PSP) and Corticobasal Degeneration (CBD) according to the latest consensus criteria (15), Dementia with Lewy Bodies (DLB) based on the 2017 revised criteria (16), and Vascular Parkinsonism (VaP) using established clinical-radiological criteria (17). The PS group was heterogeneous, comprising multiple system atrophy (MSA), progressive supranuclear palsy (PSP), and other related disorders, with a detailed breakdown provided in Supplementary Table S1. The “group” variable in the dataset, reflecting this final diagnostic classification, was used as the outcome for this study.

Patients were selected for the final analysis based on the availability of complete data for the variables included in the statistical model. Specifically, as per the pre-defined analysis plan (“SPECT-missing patients were excluded”), patients with missing or uninterpretable SPECT data (essential for categorization into “no uptake,” “uniform,” or “non-uniform”) were excluded from this cohort of 173 patients. The study protocol was approved by the Institutional Review Board of The First Hospital of Hebei Medical University, and the requirement for informed consent was waived due to the retrospective nature of the study and the use of de-identified data. This study adheres to the principles of the Declaration of Helsinki and relevant STARD (Standards for Reporting of Diagnostic Accuracy Studies) guidelines (18).

The following data were extracted from patient medical records for the analyzed cohort:

Demographic and clinical data: Age at examination (years), sex, disease duration (months from symptom onset to examination), UPDRS-III score, and Hoehn and Yahr (H-Y) stage. Blank entries for UPDRS-III or H-Y in the raw dataset were treated as missing for those specific entries but patients were included if other core data were present. Comorbidities such as hypertension, diabetes, and history of cerebral infarction were recorded as present or absent.

Non-motor symptoms: Constipation and Hyposmia (Smell Abnormality) were documented as present or absence based on patient history or clinical assessment.

131I-MIBG scintigraphy: Cardiac MIBG scintigraphy was performed according to a standardized institutional protocol consistent with general guidelines (9). Patients received an intravenous injection of 185 MBq of ¹³¹I-MIBG after thyroid blockade. Planar thoracic images were acquired at approximately 15 min (early) and 4 h (delayed) post-injection. The heart-to-mediastinum (H/M) ratio was calculated from ROIs placed on the heart and upper mediastinum. To ensure consistency, a semi-automated method was used, where ROIs were initially placed by software and subsequently reviewed and manually adjusted if necessary by an experienced nuclear medicine physician (P. Y.) blinded to the clinical diagnosis. The 4-h clearance rate was calculated as: ([Early H/M - Delayed H/M] / Early H/M) x 100%. The MIBG uptake quality was noted, and patients with non-diagnostic MIBG scans were excluded.

SPECT imaging: SPECT findings were qualitatively interpreted by experienced nuclear medicine physicians blinded to the final clinical diagnosis. For the purpose of this study, SPECT results were categorized as: No uptake (Code 0): Severe, bilateral reduction of tracer uptake. Uniform (Code 1): Generally symmetrical and relatively preserved or mildly symmetrically reduced uptake. Non-uniform (Code 2): Asymmetric reduction in tracer uptake, typically more pronounced in the putamen contralateral to the more affected limb, or a clear asymmetric pattern.

Statistical analyses were performed using SAS software, version 9.4 (SAS Institute Inc., Cary, NC, United States). Continuous variables were expressed as median and interquartile range (IQR), and categorical variables as number (n) and percentage (%). Comparisons between the PD and PS groups were made using the Wilcoxon two-sample test for continuous variables and the Chi-square test or Fisher’s exact test for categorical variables, as appropriate. A two-sided p-value < 0.05 was considered statistically significant.

The variable selection process was conducted in two stages. First, candidate predictors were chosen based on their established diagnostic relevance in the literature and clinical availability. These included demographic data (age), clinical metrics (disease duration, UPDRS-III, H-Y stage), key non-motor symptoms (constipation, hyposmia), and all available MIBG and SPECT imaging parameters. In the second stage, these candidate predictors were subjected to univariate analysis. Variables demonstrating a potential association with the diagnosis (defined by a p-value < 0.1) were then included in the multivariate logistic regression model. A backward stepwise selection method was used, with p-values for entry and removal set at 0.1. The model’s discrimination was assessed using the area under the ROC curve (AUC), and its calibration was assessed using the Hosmer-Lemeshow test.

Based on the final logistic regression model, a simplified integer-based scoring system was developed. Coefficients from the logistic regression model were used to guide the assignment of points for different levels or categories of the retained predictor variables, aiming for clinical ease of use (19).

The diagnostic performance of individual MIBG parameters, SPECT findings, the logistic regression model, and the derived scoring system was evaluated using receiver operating characteristic (ROC) curve analysis. The AUC with its 95% confidence interval (CI) was calculated. The optimal cutoff value was determined using Youden’s index. The DeLong test was used to compare the AUCs of different models, with the combined logistic model serving as the reference.

A total of 173 patients formed the study cohort, comprising 102 patients with a final diagnosis of PD and 71 patients with PS. Baseline demographic and clinical characteristics of these two groups are summarized in Table 1. Patients in the PD group had a slightly longer disease duration compared to the PS group (median 29 months vs. 24 months, p = 0.073). Significant differences were observed in MIBG scintigraphy parameters: the PD group had significantly lower 15-min H/M ratios (p < 0.001) and 4-h H/M ratios (p < 0.001) compared to the PS group. The 4-h MIBG clearance rate was significantly different between groups (median 0.02 in PD vs. −0.08 in PS, p = 0.001). SPECT findings also differed significantly between the groups (p < 0.001), with a higher proportion of PD patients showing non-uniform uptake (72.55%) compared to PS patients (47.89%).

Variables with p < 0.1 in univariate analysis were entered into the logistic regression model. After backward selection, the final model retained 4-h H/M ratio, 4-h clearance rate, and SPECT findings as significant independent predictors of PD (Table 2). The Hosmer-Lemeshow test indicated good model calibration (χ² = 8.12, p = 0.422), suggesting no significant difference between the predicted probabilities and observed outcomes. The calibration of the model is further illustrated in Supplementary Figure S1. Lower 4-h H/M ratios and non-uniform SPECT patterns were associated with an increased likelihood of a PD diagnosis. Conversely, a higher 4-h clearance rate was associated with a higher likelihood of PD.

A simplified scoring system, termed the “Combined Score,” was developed based on the variables retained in the logistic regression model. Points were assigned to different categories or ranges of these variables to create an additive score, as detailed in Table 3. This table provides the complete scoring algorithm. The total score could range from 3 to 9 points, with higher scores designed to indicate a higher likelihood of PD.

The diagnostic performance of individual parameters, the logistic regression model, and the Combined Score are detailed in Table 4 and Figure 1. The 4-h H/M ratio showed good individual predictive ability with an AUC of 0.766. The combined logistic regression model demonstrated the highest diagnostic accuracy with an AUC of 0.810 (95% CI 0.744–0.876), achieving a sensitivity of 81.4% and specificity of 70.4%. The simplified Combined Score also showed strong diagnostic performance with an AUC of 0.800 (95% CI 0.736–0.864), which was not significantly different from the logistic model’s AUC (p = 0.563). At an optimal cutoff value of ≥5 points, the Combined Score yielded a sensitivity of 78.4% and a specificity of 70.4%.