OBJECTIVES: The literature provides little guidance on statistical methods for estimating parameters of Markov models using longitudinal data. We compared the commonly used naïve (based on raw data) and advanced approaches to estimate two model parameters: transition probabilities and hospitalisation rates. Both the naïve and advanced approaches were applied using data from the European Schizophrenia Cohort (EuroSC) to populate a Markov model in schizophrenia. METHODS: EuroSC is a 2-year observational study of patients with schizophrenia (n=1,208), with 5 visits at 6-month intervals. Patients were classified into 8 health states at each visit according to severity of symptoms assessed using the Positive and Negative Syndrome Scale (PANSS). For each health state, both model parameters (hospitalisation days and transition probabilities) were estimated based on raw data by pooling all time intervals (i.e. naïve approach). Similarly, for advanced methods, transition probabilities were estimated using multi-state models while hospitalisation days were estimated using two-part Generalised Estimating Equations (GEEs). Advanced methods adjusted for patient characteristics and included random effects to account for repeated measures. RESULTS: The naïve approach showed that the average number of hospitalisation days in a 6-month interval ranged from 4.20 in health state 1 to 19.43 in health state 8. Results from the two-part GEEs provided a range from 4.21 in health state 1 to 14.7 in health state 8. GEEs tended to provide narrower confidence intervals. With regards to transition probabilities, differences between the naïve approach and the multi-state model were mostly seen in the second decimal place. CONCLUSIONS: While the naïve approach is frequently used for its simplicity, it has a number of shortcomings including: not accounting for repeated measures and not allowing for adjustment of patient characteristics. To increase the robustness of results, we recommend using statistical models that recognise and account for the unique distributional characteristics of data.