All Times EDT
Keywords: RWD, RWE, Imaging, deep learning
While there is much promise in the use of real world data to form insights on medications, comparisons of patients treated in general clinical practice with those studied within a clinical trial setting remains a challenge. One key difference is that eligibility criteria for clinical trials often includes subjective measures of patient prognosis that are not captured within the electronic health record in the real world setting. Another limitation is that specific clinical trial procedures for determining response and progression to assess a new therapy are not routinely collected by treating oncologists in general clinical practice. New metrics derived directly from tumor images may be used to assess patient prognosis prior to start of therapy as well as response determination post-therapy. We have developed a fully automated approach based on deep learning to accurately identify, segment, and quantify the volume of metabolically active tumors from FDG-PET/CT scans obtained from DLBCL and FL patients who participated in clinical trials. We show how this can be the basis of endpoints and prognostic markers calculated in an analogous way for patients enrolled in clinical trials and patients being treated in the real world setting. We discuss the opportunities this presents for generating insights for new cancer treatments.