Development and Validation of a Prognostic Model for Cognitive Impairment in Parkinson's Disease With REM Sleep Behavior Disorder

Abstract

The presentation and progression of Parkinson's disease (PD) are not uniform, but the presence of rapid eye movement sleep behavior disorder (RBD) in PD patients may indicate a worse prognosis than isolated PD. Increasing evidence suggests that patients with comorbid PD and RBD (PD-RBD) are more likely to develop cognitive impairment (CI) than those with isolated PD; however, the predictors of CI in PD-RBD patients are not well understood. This study aimed to develop a prognostic model for predicting mild cognitive impairment (MCI) in PD-RBD patients. The data of PD-RBD patients were extracted from the Parkinson's Progression Markers Initiative study (PPMI), and the sample was randomly divided into a training set (n = 96) and a validation set (n = 24). PD-MCI as defined by the level II Movement Disorder Society (MDS) diagnostic criteria was the outcome of interest. The demographic features, clinical assessments, dopamine transporter (DAT) imaging data, cerebrospinal fluid (CSF) analyses and genetic data of PD patients were considered candidate predictors. We found that performance on the University of Pennsylvania Smell Identification Test (UPSIT), the mean signal and asymmetry index of the putamen on DAT imaging, p-tau/α-syn and p-tau in CSF, and rs55785911 genotype were predictors of PD-MCI in PD-RBD patients. A C-index of 0.81 was obtained with this model, and a C-index of 0.73 was obtained in the validation set. Favorable results of calibrations and decision curve analysis demonstrated the efficacy and feasibility of this model. In conclusion, we developed a prognostic model for predicting MCI in PD-RBD patients; the model displayed good discrimination and calibration and may be a convenient tool for clinical application. Larger samples and external validation sets are needed to validate this model.

Keywords: Parkinson’s disease; REM sleep behavior disorder (RBD); cognition; mild cognitive impairment; nomogram.

Figure 1

Establishment of the prediction model. (A) Kaplan–Meier plots of mild cognitive impairment (MCI)-free survival between Parkinson’s Disease (PD)-RBD patients and PD-nRBD patients. (B) Optimal parameter selection in the least absolute shrinkage and selection operator (LASSO) regression model. (C) The coefficient profiles of 83 variables in the LASSO model. The lambda we selected resulted in six variables with nonzero coefficients. (D) ROC curves of model 1 and model 2 for predicting MCI in PD-RBD patients at 3 years. (E) ROC curves of model 1 and model 2 for predicting MCI in PD-RBD patients at 5 years. (F) ROC curves of model 1 and model 2 for predicting MCI in PD-RBD patients at 7 years.

Figure 2

Flow diagram of patient selection and allocation (A) and nomogram for predicting the MCI-free probabilities at 3, 5, and 7 years in PD-RBD patients (B). Each level of every predictor corresponds to a score on the points scale, and the total points were acquired by adding the scores of each variable. The total points for each patient correspond to the estimation of the 3-, 5-, and 7-year probabilities of MCI-free survival in this nomogram. Abbreviations: PD, Parkinson’s disease; RBD, Rapid eye movement sleep behavior disorder; MCI, mild cognitive impairment; ai, asymmetry index.

Figure 3

Calibration plots of model-predicted MCI-free survival probabilities at 3 years, 5 years, and 7 years in PD-RBD patients. Model-predicted probabilities of overall survival are plotted on the x-axis; actual overall survival is plotted on the y-axis. (A) The calibration curve of model 1-predicted MCI-free survival probabilities at 3 years, 5 years, and 7 years in the training set, n = 96. (B) The calibration curve of the model 2-predicted probability of MCI-free survival probabilities at 3 years, 5 years, and 7 years in the training set, n = 96.

Figure 4

Time-dependent AUC and decision curve analysis (DCA) for model 1 and model 2. (A) Time-dependent AUC plot showing how the AUCs of model 1 and model 2 changed with follow-up time. (B) DCA of model 1 and model 2 for 3-, 5-, and 7-year predictions. The brown line represents the assumption that all PD-RBD patients have MCI. DCA showed the following: (1) all patients would benefit from model 1 and model 2 for 3-year prediction; (2) for 5-year prediction, patients at risk threshold <0.75 would benefit from model 2, while patients at almost all risk thresholds would gain a positive net benefit in model 1; and (3) using model 1 and model 2 benefits patients more than the treat-all scheme or the treat-none scheme at almost all risk thresholds.

Figure 5

Calibration plots of model 1-predicted probabilities of MCI-free survival at 3 years, 5 years, and 7 years in the validation set. Model-predicted probabilities of overall survival are plotted on the x-axis; actual overall survival is plotted on the y-axis. (A) The calibration curve of the model 1-predicted probability of MCI-free survival at 3 years in the validation set, n = 24. (B) The calibration curve of the model 1-predicted probability of MCI-free survival at 5 years in the validation set, n = 24. (C) The calibration curve of the model 1-predicted probability of MCI-free survival at 7 years in the validation set, n = 24.