OBJECTIVES: In the Netherlands, a new drug is priced to the level of an existing drug cluster, if these are considered similar. However, when a drug is not considered similar to a cluster, a separate price can be requested. Before this separate price is approved, sufficient cost-effectiveness and budget impact (BI) data have to be provided to the Dutch Health Care Insurance Board (CVZ). The present analysis investigated how close the submitted BI-forecasts were to the actual BIs for cancer treatments in the Netherlands. METHODS: The publicly available evaluations by the CVZ were assessed for cancer agents. From these, forecasted number of users, drug costs, treatment lengths and the BIs were derived. The predicted BIs were compared with actual BIs from the GIP-database (Dutch Drug Information System). The predicted BI was standardized with respect to time to ensure that actual and forecasted BIs shared the same starting point. Broadening of the indication was taken into account and substitution was considered. To further explain any difference in forecasted and actual BI, various variables such as drug cost, number of users, market share (when applicable) and treatment lengths were investigated. RESULTS: The search provided five relevant cases. The forecasted budgets were lower than the actual ones in 4 out of the 5 cases. The forecasted and actual BI differed up to 250%. Data on drug substitution was insufficient and therefore not considered. The differences between predicted and observed BIs were explained primarily by an underestimation of the number of patients eventually receiving the evaluated treatments. The published BI forecasts contained no/very limited sensitivity analyses. CONCLUSIONS: Most BI forecasts underestimated the actual BI for new cancer drugs. One explanation was that the growth of the patient population was often underestimated. Improvements in predicting total market-size and -penetration should be considered, as well as more elaborate sensitivity analysis.