UCLA David Geffen School of Medicine/UCLA Medical Center Los Angeles, CA
M. A. Eala1, J. Hernandez1, N. Chong1, M. L. Steinberg1, A. Kalbasi2, V. Reddy1, and R. R. Savjani1; 1Department of Radiation Oncology, University of California, Los Angeles, Los Angeles, CA, 2Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA
Purpose/Objective(s):In prospective clinical trials, data from enrolled patients are meticulously kept for each patient by the study investigators. After trial completion, these data are distilled into a few static figures that describe the primary and secondary objectives and endpoints of the study. However, large portions of the trial data are never examined as they remain siloed in tables that are largely inaccessible to both the study investigators and to oncology researchers broadly. We showcase how an online, interactive data visualizer can be built and deployed to allow for rapid data visualization of any variables collected as part of a clinical investigation. Materials/
Methods: We built a data visualization tool using all data from a completed phase II clinical trial in our department on pre-operative 5-day radiotherapy for soft tissue extremity sarcomas. Data were stored in our institutional database, which allowed for easy export using an API with a token. Importantly, data were anonymized such that no PHI information were included to identify patients. Using python packages for plotting Plotly and Dash, we constructed an online dashboard that allowed interrogation of all clinical trial variables for population inference, including demographic data, clinical outcomes, pathology results, surgical complications, treatment toxicities, and images from radiotherapy planning. Results: The online dashboard allowed users to dynamically select any variables to plot against each other or in groups. This included scatter plots with customizable sizes and colors of data on selected variables, histograms grouped by any categorical variable, box and scatter plots, survival plots for local failure/distant failure/overall survival (with integrated competing risk analyses), swimmer’s plots for wound complications, and a geographic map for patient enrollment. Further, we generated a tab that allowed for visualization of the gross tumor volume and radiation dose map overlayed on the planning CT Simulation scan in a montage. Lastly, we included a table to allow dynamic patient filtering by selecting any combination of criteria (e.g., age > 70 AND tumor size > 10 cm). The dashboard automatically updated to include only the selected patients. We deployed our dashboard online via the Google Cloud Platform. Conclusion: Our interactive data visualizer allows all data from a clinical trial to be readily accessed. This approach puts warehoused data into tables now at the fingertips of all interested clinicians and researchers without needing to request access, write additional code, or utilize any additional software to visualize data. We believe this novel approach should transform how clinical trial data are shared to enhance transparency and help discover otherwise unseen insights in clinical trials.
