Characterizing Side Effects of Lung Cancer Targeted Therapies via High-Throughput Analysis of Electronic Health Record Databases

Abstract

Lung cancer is the second most common cancer in both men and women in the United States and the leading cause of cancer death. Targeted therapies, which represent an effective treatment option for the disease, work by inhibiting specific cancer driver mutations identified via genetic sequencing of the tumor. Nearly 30 such drugs are approved for standard-of-care use with many more under development. Despite the prevalence of this class of therapies, collecting generalizable information on individual drug toxicities remains challenging given the infrequent use of most targeted therapies by single treatment centers. The CancerLinQ Discovery database synthesizes and deidentifies electronic health record data from millions of cancer patients treated at a variety of practice sizes and geographic distributions. Diagnosis codes present in the database not only provide cancer diagnosis details but can also be used to understand side effects of specific treatments. In this project, I leverage deep phenotyping, high-throughput association studies, and time-to-event analysis to characterize the side effect profiles of lung cancer targeted therapies from the CancerLinQ database. High-throughput phenome-wide association studies applied to this database not only identify clinically significant side effects concordant with the published toxicity profiles of lung cancer targeted therapies, but also identify unique side effects underreported in existing post-marketing surveillance databases that warrant further investigation. Time-to-event analyses further expand upon this toxicity characterization by incorporating details about the timing of specific side effects. This framework scales across diverse medical therapies and integrates well with other informatics tools such as clinical decision support and patient-reported outcome research.