Extracting structured information from free text pathology reports

We have developed a method that extracts structured information about specimens and their related findings in free-text surgical pathology reports. Our method uses regular expressions that drive a state-automaton on top of XSLT and Java. Text fragments identified are coded against the UMLS. This paper describes the technical approach and reports on a preliminary evaluation study, designed to guide further development. We found that of 275 reviewed reports, 91% were coded at least so that all specimens and their critical pathologic findings were represented in codes.