Information Extraction from Diagnostic Reports to Improve Patient Recruitment Efforts for Lung Cancer Clinical Trials

Recruiting patients for a clinical trial is an arduous and time-consuming task involving significant manual efforts and disparate sources of information available in patients' medical records. On the other hand, inability to recruit in a timely and unbiased fashion has been the major hurdle for academic researchers and pharmaceutical companies in developing new treatment options. Due to this, automated eligibility screening approaches are becoming increasingly popular to alert healthcare providers of potential eligible patients. These approaches rely on a representation of the trial criteria that is typically checked against information in both structured and textual components of EMRs. This project focuses on automatic extraction of positive lung cancer diagnoses and the associated details of disease progression and observed metastases from radiology reports. The extraction is modeled as a prediction problem using machine learning methods. To demonstrate the value of this information, an initial prototype will be built that employs both the extracted information and other elements of interest from structured sources from MCC and KCR to rank potential eligible cases and eventually generate alerts based on this ranking.