As there is no data in the literature about maximum survival, 14 years was chosen to represent the maximum

lifetime of a patient (less than 1% of patients were alive in the model at 14 years). Due to the chronic, progressive, and evolutionary nature of the disease, a Markov modeling approach was employed, following patients as they selleck compound passed through a series of clearly defined and mutually exclusive health states throughout their disease. The model was designed to track the health states of patients with HCC in both treatment arms. Patients received first-line treatment (sorafenib or BSC) until documentation of disease progression or until a treatment-limiting AE occurred (first-line treatment—no progression). At the point of progression, patients could either continue on first-line treatment (first-line treatment continued—post-progression) or switch to BSC (BSC—post-progression). At any point in the model, patients could die due to all-cause (general) mortality (Fig. 1). The model structure is consistent with clinical practice and other economic models developed in oncology,14–18 and was validated by clinical experts in the USA. The model used monthly cycles (30 days) to match treatment patterns, that is, patients have the possibility to move from one health state to another every month. For the analysis, the following assumptions

were made: The HCC population and the efficacy data from the SHARP trial were generalizable to the USA; Health effects are expressed as life-years (LY). Costs are given as 2007 US dollars. Results are presented as incremental LY, incremental Palbociclib cost costs, and incremental cost-effectiveness ratios (ICER) in terms of cost per LY gained. An annual discount rate of 3%19 was applied to both costs and health benefits occurring beyond the first year. The model used the effectiveness data from the SHARP trial. These

included TTP according to investigator radiological assessment, OS, sorafenib dosing, and the rate of all grade 3 and grade 4 AE. Due to the statistically and clinically significant OS benefit observed at an interim analysis in the sorafenib arm, the SHARP trial find more was stopped at 72 weeks. Therefore, the OS and TTP results were extrapolated using survival functions that best fit the patient-level data. (Calculations were done in stata). Assuming nothing else changes in the two treatment arms except time, these estimated survival functions utilize all available data and thereby lead to the most accurate extrapolation.20 The Akaike information criteria, which measure the goodness-of-fit of an estimated statistical model,21 showed that a lognormal model provided a significantly better fit compared to a Weibull, loglogistic, exponential, or Gompertz distribution in the sorafenib subgroup, and an equally good fit as the loglogistic distribution in the BSC subgroup. Thus, lognormal distribution was chosen (Fig. 2).