Progression of Parkinson's disease (PD) can be detected through changes in clinical ratings or disability assessments. A clinical trial, Deprenyl and Alpha-Tocopherol Antioxidative Therapy of Parkinsonism (DATATOP), used a novel study end point: increase in parkinsonian disability to the extent that investigators determined the need for treatment with levodopa. We analyzed DATATOP results to learn if this operationally defined end point could be reproduced from elements of the Unified PD Rating Scale (UPDRS) and other conventional clinical scales. Our analysis involved UPDRS, Schwab and England Activities of Daily Living (S-E ADL), and Hoehn and Yahr (H-Y) scores when DATATOP subjects reached the study end point. Various UPDRS components were examined, including subscores measuring severity of impaired ADL, bradykinesia, postural instability and gait difficulty, tremor, and rigidity. Data from subjects reaching the end point were compared with assessments of those DATATOP subjects who did not, matched for the same duration of enrollment. All measures showed subjects who reached the end point had significantly greater mean impairment than did controls, although the two groups had substantial overlap. Multivariate analysis by using conditional logistic regression suggested that the end point was determined more by functional (S-E ADL and the UPDRS ADL scores) than by clinical examination criteria. The method of classification and regression trees suggested a simple decision tree splitting, respectively, on S-E ADL, UPDRS ADL, H-Y score, and UPDRS ADL again, with an estimated overall misclassification probability of 18%. We conclude that the DATATOP end point cannot be fully reproduced from the traditional clinical measures, although it can give results that are consistent with these scales in a well-designed clinical trial.