Dynamic Prediction of Excessive Daytime Sleepiness Through Random Survival Forest: An Application of the PPMI Data

Abstract

Parkinson’s disease (PD) is the second most widespread neurodegenerative disease worldwide. Excessive daytime sleepiness (EDS) significantly correlates with de novo PD patients. Identifying predictors is critical for the early detection of disease. We investigated clinical and biological markers related to time-dependent variables in sleepiness for early detection of PD. Data were obtained from the Parkinson’s Progression Markers Initiative study, which evaluates the progression markers in patients. The dataset also includes various longitudinal endogenous predictors. The measures of EDS were obtained through the Epworth Sleepiness Scale (ESS). The random survival forest method, which can deal with multivariate longitudinal endogenous predictors, was used to predict the probability of having EDS in PD. The rate of having EDS among PD patients was 0.452. The OOB rate was 0.186. The VIMP and minimal depth indicated that the most important variables are stai state, JLO, and the presence of the ApoE4 Allele. In early PD, EDS is a good indicator of the diagnosis of the PD and it increases over time and has associations with several predictors