2.1. Study design
Study data used in the present study were obtained from the Parkinson's Progression Markers Initiative (PPMI) database (www.ppmi-info.org/data). PPMI is an ongoing observational, international, multicentre cohort study aiming to identify blood-based, genetic, spinal fluid, and imaging biomarkers of Parkinson’s disease (PD) progression with longitudinal follow-up in a large cohort. The aims and methodology of PPMI study have been published elsewhere17. Study protocol and manuals and are available online. The study was approved by the Institutional Review Board at each site, and all participants provided written informed consent.
For this study, we utilized the baseline dataset of PPMI from 33 participating outpatient PD treatment centers worldwide on the basis of inclusion and exclusion criteria previously published. All the methods were performed in accordance with relevant institutional guidelines and regulations. The Strengthening the Reporting of Observational studies in Epidemiology (STROBE) guideline was included in supplementary material 1. Figure 1 illustrated the selection process of our study. Participants who were diagnosed as idiopathic PD, with baseline Rapid eye movement (REM) sleep behavior disorder questionnaire (RBDSQ) score and BMI data were included as study population. Healthy controls, scans without evidence of dopaminergic deficit (SWEDD) patients, prodromal patients, duplicated participants, individuals with missing baseline RBDSQ score or BMI data were excluded from this analysis. In total, 1115 patients with complete information, enrolled between November 2010 and June 2023, were included in our analyses. The data were downloaded on September 15, 2023.
2.2. Exposure and outcome
Exposure was assigned as baseline BMI. Anthropometric data, including height and weight, were obtained from baseline data. BMI was calculated as weight divided by height squared (weight(kg)/height(m2)), and then classified into 4 WHO categories, including underweight (BMI < 18.5 kg/m2), normal weight (18.5 ≤ BMI < 25.0 kg/m2), overweight (25 ≤ BMI < 30 kg/m2) and obese (BMI ≥ 30 kg/m2) categories18,19.
Outcome was assigned as REM-sleep behavior disorder (RBD) at baseline. The 10-item RBDSQ has been validated in PD patients and demonstrates good accuracy in identifying RBD. Items 1 to 4 of the RBDSQ assess the frequency and content of dreams, as well as their relationship to nocturnal movements and behavior. Item 5 asks about self-injuries and injuries to the bed partner. Item 6 consists of four subitems specifically assessing nocturnal motor behavior, including questions about nocturnal vocalization, sudden limb movements, complex movements, or items falling from the bed. Items 7 and 8 inquire about nocturnal awakenings. Item 9 focuses on general sleep disturbances, while item 10 pertains to the presence of any neurological disorder. The maximum total score on the RBDSQ is 13 points. Following the definition set by the International Parkinson and Movement Disorders Society (MDS) Task Force, we defined RBD as a baseline RBDSQ score equal to or greater than 520.
2.3. Covariates
To assess the potential influence of confounding factors, several important covariates were selected as a prior based on the literature. These covariates included age, sex, PD duration, depression (measured by the 15-item Geriatric Depression Scale [GDS-15] score in PPMI), levodopa equivalent daily dose (LEDD), hypertension (defined as self-reported hypertension, systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, or use of antihypertensive drugs), and MDS Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) score at baseline21. All participants in the PPMI study underwent the standard test battery of assessments. In addition to the covariates mentioned above, sociodemographic characteristics and clinical battery relevant to this study including education, Hoehn & Yahr stage, serum uric acid, autonomic function assessed by the Scale for Outcomes for Parkinson’s Disease-autonomic function (SCOPA-AUT) score, anxiety assessed by State-Trait Anxiety Inventory (STAI) score, and daily living quality assessed by Modified Schwab and England Activities of Daily Living Scale (MSEADL) were adjusted as potential confounders in the models. Although PPMI collected an array of cerebrospinal fluid (CSF) biomarkers, these measures were only available for a small subset of participants and thus were not included in this study.
2.4. Statistical analysis
Summary statistics were performed and tested for normality (Shapiro-Wilk). Continuous data were presented as median (interquartile range [IQR]) or mean ± standard deviation (mean ± SD), with categorical data presented as proportion and number (N [%]) as appropriate. Group comparisons were analyzed with use of Student’s t tests or Wilcoxon’s rank-sum tests for continuous data and Chi-square tests or Fisher’s exact tests for categorical data. Data were more than 99% complete. The remaining missing values were imputed by multivariable chained imputation with fully conditional specification, and imputed and reported results were similar22. All statistical tests were two-sided and the level of significance was set at 0.05.
Participants were divided into two groups based on whether they had RBD. The relationship between BMI and RBD was examined using logistic regression models. The adjustment was accomplished via 3 models: (1) model 1, without any covariate adjustment; (2) model 2, adjusted for age, sex, PD duration, GDS-15 score, LEDD, hypertension, depression, and MDS-UPDRS I, II and III scores; (3) model 3, additionally adjusted for education level, serum uric acid, Hoehn & Yahr stage, STAI, SCOPA-AUT, and MSEADL scores as covariates. The results were presented as odds ratios (OR) with corresponding 95% confidence intervals (CI).
Restricted cubic spline (RCS) analysis was also performed to examine the association of between baseline BMI and RBD based on univariable and multivariable adjusted logistic regression models23. To balance best fit and overfitting in the main splines, the number of knots, between three and seven, was chosen as the lowest value for the Akaike information criterion. If the difference in the number of knots was within two for different models, the lowest number of knots was selected24. The same number of knots from the main splines was also applied for stratified analyses to allow direct comparison of overall and stratified analyses, including test of interaction. We tested for potential non-linearity by using a likelihood ratio test comparing the model with only a linear term against the model with linear and cubic spline terms. Piecewise-linear models were then fitted to quantify the association between BMI and RBD. If evidence of non-linearity was found, a two-line piecewise linear model with a single change point was estimated by trying all possible values for the change point and selecting the value with the highest likelihood among those considered, while accounting for covariates.
We fitted interactions to investigate effect modification by depression (with depression, without depression, GDS-15 ≥ 5 or not), motor subtype (tremor dominate [TD], non-TD including postural instability/gait disorder [PIGD] or Indeterminate), sex (male, female), and hypertension (yes, no) 25–27. Due to the nonlinear association between BMI and RBD in the whole participants, we used continuous BMI and the quadratic term BMI2 in multivariable adjusted logistic regression models (model 3) to allow for the nonlinearity during the interaction analyses. The first model to test for the depression-by-BMI interaction allowed for interaction with both the linear and quadratic terms of BMI28.
Model A: RBD = BMI + BMI2 + depression + BMI × depression + BMI2 × depression + other covariates
In the absence of interaction with the quadratic term, the model was then simplified to only allow for interaction with the linear term.
Model B: CVD = BMI + BMI2 + subtype + BMI × depression + other covariates
The significance of the interaction was determined based on the highest level interaction term in the models, and here, lack of interaction was inferred when neither BMI2 × depression (Model A) nor BMI × depression (Model B) were significant at the 5% level. Interactions by subtype, hypertension, and sex were examined in the corresponding manner, replacing “depression” above with hypertension, sex or subtype as relevant.
To assess the robustness of the results, we additionally applied three sensitivity analyses. First, we examined the shape of BMI-RBD relation after excluding individuals who had a baseline Montreal Cognitive Assessment (MOCA) score less than 26 as the definition of cognitive impairment29. Second, we restricted the analysis to individuals in the first and second categories of education. Third, we performed the analysis after excluding other ethnic groups. All data were analyzed using R (version 4.0.2).