Objective: This study evaluated natural language processing methods to extract clinical data from free text in surgical reports related to cardiac pacing and defibrillation in order to populate a registry.
Methods: The information extraction system that we have developed is a name entity recognition system based on GATE using regular expressions. 232 reports were analyzed. For each report, we performed manual abstraction, we collected the information stored in the registry, and we performed information extraction with our system. Sensitivity,positive predictive value and accuracy were used to evaluate our method.
Results: Our system extracted information, including numeric data, text and combination of numbers and strings, with a high sensitivity (>90%) and very high predictive positive value (>95%). It featured a precision that was higher than the precision of the original registry database populated by manual input.Conclusion This tool based on GATE open source components provides a robust approach to extracting information from documents related to a specific narrow domain such as pacemaker reports. Further evaluation is needed for application to broader domains.