OBJECTIVES: Parkinson’s disease (PD) is a frequent and chronic neurological disorder representing a growing burden for national health care budgets. We assessed independent predictors for 1) PD-specific drug costs; 2) direct non-drug costs'; and 3) indirect costs METHODS: We analyzed data from an ongoing prospective cost study (n=157) within the German PD Competence Network using multivariate regression techniques. Potential predictors included sociodemographic factors, clinical data represented by the Unified Parkinson’s Disease Rating Scale (UPDRS), disease severity (Hoehn&Yahr scale) and quality of life (QoL) data. PD-specific drug costs and non-drug costs were log-transformed prior to model development, resulting in multiplicative models. Indirect costs were calculated using the human capital approach. Model formulation proceeded in two steps: first, presence of indirect costs was modeled by logistic regression, and second, costs were predicted by linear regression in the subgroup of patients with indirect costs. RESULTS: Predictors of PD-specific drugs included age, sex, clinical state (all p<0.001) and QoL (p=0.02). Non-drug direct costs were predicted by disease stage p=0.05 and p<0.001) and QoL (p=0.03). Presence of indirect costs depended on age (p<0.001), UPDRS (p=0.03), QoL (p=0.04), presence of depression (p=0.02), and falls (p=0.006). The magnitude of indirect costs was predicted by UPDRS (p=0.003) and falls (p=0.007). Variance explained by the predictors (adjusted R-squared) ranged from 24% to 28% for all models. CONCLUSIONS: Based on our analysis, clinical and QoL variables are the most relevant predictors of PD cost. However, we cannot infer from this study whether improved QoL has an effect on PD costs, or is a consequence of more expensive treatment.