The National Health Registries managed by the Institute of Health Information and Statistics (ÚZIS) are a healthcare information system used for the implementation of national health policies and for medical and public health research. One of these registries is the National Oncology Registry (NOR), which is used to centralize and manage data about cancer patients. This registry is filled by a range of organizations, from healthcare facilities to health insurance companies. The aim of the work is to train a machine learning system that will enable the analysis of data in the NOR, which are provided by individual organizations, and to identify the most relevant records in overlapping entries. The system will return the most relevant record, or possibly a probabilistic evaluation of each record. The system will be trained on data evaluated by experts between the years 2019-2021 and will be verified on expertly resolved data from 2022 (to the extent possible - otherwise testing data needs to be split from the 2019-2021 data). The work will include experimental evaluation and comparison of different models using metrics commonly used in machine learning. Furthermore, the models will be analyzed from the perspective of explainability - either through their own explainability or by using post-hoc methods. The system will be implemented in a project collaboration with ÚZIS.