OBJECTIVES:  The Unified Parkinson's Disease Rating Scale (UPDRS) is a widely used clinical rating scale in Parkinson’s disease (PD), and well recognized in reflecting the disease’s multifactorial impact. In the absence of EQ-5d data in clinical studies, mapping from disease-specific questionnaires is a valid alternative. We aimed to develop a mapping algorithm from UPDRS to EQ-5D-3L utilities for the UK, based on data from the EARLYSTIM study, a randomized controlled trial of Deep Brain Stimulation(DBS) in PD with early motor complications (Schupbach et al. 2013). METHODS: A two-step approach was applied, based on following EARLYSTIM data (baseline and follow-up): UPDRS I-IV, PDQ-39, age, gender.1^st step: A published multinomial logistic regression model (Kent 2015) was applied to map PDQ-39 to EQ-5D-3L. 2^nd step:UPDRS, gender, and age were paired with EQ-5D-3L values as a basis for development of the mapping algorithm (732 paired observations). A range of statistical models (linear/beta regression, finite mixtures of linear/beta regression) with various link functions were developed. Models were compared for prediction accuracy (e.g.mean error (ME), mean absolute error (MAE), Bayesian Information Criterion (BIC))and internal validity. RESULTS: The differences in prediction accuracy were small for all models considered. The finite mixture models had slightly better performance than the single equation regression models. Three models showed best performance and internal validity, namely: i)beta regression with log link function model, ii)two-component finite mixture of beta regressions with cloglog link function model, and iii)three-component finite mixture of beta regressions with log link function model. All 3 algorithms contained coefficients for UPDRS parts I, II, III (except beta regression model), IV, age and gender. CONCLUSIONS: The EQ-5d mapping algorithms developed from UPDRS provide an opportunity for use of clinical outcomes data from existing PD studies in cost-effectiveness evaluations. Further research should include applying the algorithms to a different data set for confirmation of external validity.