Fig. 2: The progression patterns of Parkinson’s disease subtypes in the discovery dataset.
From: Two distinct trajectories of clinical and neurodegeneration events in Parkinson’s disease
a The positional density maps of two Parkinson’s disease subtypes. x axis indicates the number of events (11 variables * 3 grads). The color intensity reflects the row-wise positional density and confidence in the ordering. The location of highest color intensity means the most probable stage/sequence of this feature. b Distribution of two subtypes across SuStaIn stages. c Boxplots of probability of maximum likelihood subtype. d Log Likelihood across Markov chain Monte Carlo iterations. e Log Likelihood across 10-fold cross-validation. Log Likelihood increased dramatically from 1 subtype to 2 subtypes, but not increased slowly or even decreased from 2 subtypes to 3 or more subtypes. f The CVIC under different number of subtypes. Decreased CVIC illustrated improved model fit. In case of little improvement in model fit, a simpler model should be favored. Box plots with center line indicating median, bounds of boxes showing upper and lower quartile, whiskers illustrating 1.5 * interquartile range, and dots representing the distribution of raw data (minima and maxima are included). The contrast-to-noise ratio of substantia nigra and locus coeruleus, the free-water of basal forebrain, entorhinal cortex, amygdala, and hippocampus were enrolled. RBDQ-HK Rapid Eye Movement Sleep Behavior Disorder Questionnaire (Chinese University of Hong Kong version), BSIT Brief Smell Identification Test modified for Chinese, SCOPA-AUT Scales for Outcomes in Parkinson’s Disease-Autonomic, HAMD Hamilton Rating Scale for Depression, MoCA Montreal Cognitive Assessment, SuStaIn Subtype and Stage Inference, CVIC cross-validation information criterion.
