Estimations of the number of hospitalized incident cancer cases show biases when claims databases are used. This is due to false reports of incident cancer because of a lack of specificity, and because of unrecorded cancers resulting from a lack of sensitivity. We present a statistical method to provide corrected estimations. This method is based on a two-phase study design using an external data set for sensitivity and specificity estimates. Inaccuracy of the corrected number of hospitalized incident cancer cases was assessed by a credibility interval determined by a Bayesian approach using a Monte Carlo method. Based on the population hospitalized in a large group of French University hospitals, 334 women were identified in the French claims database as having potential incident cases of breast cancer in 1997. According to our method, the corrected number was 565 (550-580). In absence of hospital-based cancer registries, our approach provides estimates and credibility intervals, and has many potential applications in defining hospital policies with its applicability to other diseases.