(0331) Streamlining Clinical Data Parsing: A Robust and Generalizable Algorithmic Approach

Introduction: The age of language processing and big data mining has ushered in new techniques for extracting clinicopathologic data from the electronic medical record (EMR) and offers invaluable tools for expediting and advancing research endeavors. Here we developed and evaluated an algorithmic pipeline using regular expressions (regex) for automated extraction of various pathologic features from original text reports with ease of universal application.

Methods: Clinical pathologic reports from 1171 surgical patients who underwent resection and neck dissection in the setting of an oral cavity cancer were queried from our institutional pathology database using patient identifiers. The data were analyzed using Python, and our automated extraction was implemented using the Pandas library. Predetermined keywords were used to isolate text excerpts 300 characters before and after each keyword. Then, a regex function was used to strip the excerpt to contain only letters and numbers, and 3-5 text patterns were used to extract data of interest. The accuracy of the automated method was compared to a manual review of the dataset.

Results: We successfully extracted five variables including numeric (tumor size, nodes examined, positive nodes, and invasion depth) and character (extranodal extension) using 3-5 regex patterns each. The accuracy compared to manual reviews was high: tumor size (98.33%), invasion depth (99.74%), positive nodes (94.28%), nodes examined (93.14%), and extranodal extension (94.37%).

Conclusions: Our pipeline provides an accurate, reliable and easily customizable method for extracting variables from a complex, text-based dataset using sophisticated text patterns and regular expressions. This can be expanded to other clinicopathologic data mining to greatly enhance research endeavors across Otolaryngology.