The Act for the Protection of Personal Information in Japan considers as personal information any quasi-identifier that may be used to obtain information that identifies individuals through comparisons with datasets. Studies using health records are not widely conducted because of the concern regarding the safety of anonymized health records. To increase the safety of such records, we used the Pk-anonymity method. In this method, attributes are probabilistically randomized and then reconstructions are performed on the basis of statistical information from perturbed data. Hence, it is expected to provide more precise statistics and more reliably preserve privacy than the traditional "k-anonymity" method. We anonymized health records, performed cross tabulation, and assessed the error rate using original data. This study shows that the Pk-anonymity method can be used to perform safety statistical disclosures with low error rates, even in small cases.