OBJECTIVES: Identifying treatment pathways based on administrative data or electronic medical records is generally a complex task, as many incidental events may hide the key trends. The purpose of this study was to propose a transparent methodology for defining treatment pathways using a large longitudinal real life database. The application of two alternative algorithms with different variants in Parkinson’s disease (PD) was described. METHODS: In the first method, a new treatment line was assumed to start when a pre-specified number of consecutive prescriptions of drugs from the same PD drug class or combination of classes (rasagiline, selegiline, dopa agonists, COMT inhibitors, levodopa and derivatives) occurred. In the second method, start and end dates of continuous treatment periods (i.e. without gaps higher than six months between prescriptions) were determined for each drug class independently of each other. Combinations were assumed to be used when different treatment periods overlapped. Algorithms were tested using medical records of patients with Parkinson’s disease (PD) extracted from the UK Clinical Practice Research Datalink (CPRD). The outputs of the different algorithms were described in details for ten patients, randomly selected. The sensitivity of the algorithms to changes in assumptions was tested. RESULTS: The first algorithm did not systematically capture the addition of a new drug to the current treatment as combined drugs were not necessarily prescribed or renewed during the same consultations. However, the second method well captured the treatment lines observed but sometimes created undue treatment lines from isolated prescriptions. This issue was overcome by deleting short lines lasting less than three months. CONCLUSIONS: The second algorithm developed in this study provided an accurate description of the treatment strategies developed by prescribers in Parkinson’s disease in the UK using a limited number of assumptions and may be useful for other chronic diseases.