OBJECTIVES: To evaluate the predictive performance of a previously published mapping algorithm for converting the prostate cancer specific instrument FACT-P (Functional Assessment of Cancer Therapy–Prostate) to EQ-5D utility values (UK tariff) on a sample with a different disease stage than the one on which the model was generated [Skaltsa et. al. ViH 2014]. METHODS: We applied a previously developed algorithm to the data obtained from a randomized, double-blind, placebo-controlled phase 3 trial in asymptomatic/mildly symptomatic chemo-naïve mCRPC patients. The trial collected EQ-5D and FACT-P data at baseline and until treatment discontinuation. The mapping model was developed on mCRPC patients in a post-chemo setting, included the FACT-P subscale scores and baseline variables and used separate algorithms for patients with good and poor health defined as a FACT-P score exceeding or not 76. Model performance was assessed by mean absolute error (MAE) and root mean squared error (RMSE).  RESULTS: The testing dataset contained 1,669 patients with baseline and ≥1 post-baseline scores. The average baseline EQ-5D utility and FACT-P total score were 0.844 and 119.5 respectively. Percentage of perfect health was 37% across all visits (ceiling effect). The average (across all visits) observed and predicted EQ-5D utility index value was 0.823 and 0.842, respectively. The model yields accurate predictions (MAE=0.107; RMSE=0.150) comparable to the ones obtained on the development sample (MAE=0.117; RMSE=0.162). The model predicts well for milder health states, but overpredicts for the more severe ones (EQ-5D utility≤0.5: MAE=0.436, RMSE=0.258; EQ-5D utility > 0.5: MAE=0.096, RMSE=0.125). CONCLUSIONS: Although external validation is recommended using similar samples, our findings show that the algorithm developed in the post-chemo setting performed well in a pre-chemo setting in mCRPC patients, although overpredicts for severe states. This model seems suitable for predicting utility values for economic evaluation when a preference-based measure is absent in chemo-naïve mCRPC populations.