PURPOSE: This manuscript aims to precisely describe the natural disease progression of Parkinson's disease (PD) patients and evaluate approaches to increase the drug effect detection power. METHODS: An item response theory (IRT) longitudinal model was built to describe the natural disease progression of 423 de novo PD patients followed during 48 months while taking into account the heterogeneous nature of the MDS-UPDRS. Clinical trial simulations were then used to compare drug effect detection power from IRT and sum of item scores based analysis under different analysis endpoints and drug effects. RESULTS: The IRT longitudinal model accurately describes the evolution of patients with and without PD medications while estimating different progression rates for the subscales. When comparing analysis methods, the IRT-based one consistently provided the highest power. CONCLUSION: IRT is a powerful tool which enables to capture the heterogeneous nature of the MDS-UPDRS.