Data Science to Improve Treatment Planning for Advanced Prostate Cancer Patients Treated with Radiotherapy
Abstract
Long-term quality of life (QoL) is increasingly important for men with locally advanced prostate cancer, who can live for many years after diagnosis. Many of these men will receive radiation as part of their cancer treatment. Radiation can cause short- and long-term toxicities, including problems with urinary and bowel functioning which can negatively impact QoL, in up to 60% of patients. Improving patient QoL is a FY21 PCRP Overarching Challenge. Our team believes we can improve the lives of prostate cancer patients who receive radiation through this innovative proposal that incorporates the latest advances in artificial intelligence. The current study will address this problem using an innovative combination of novel datasets and cutting-edge data science. First, our study will incorporate large-scale patient-reported outcomes (PROs), or QoL and toxicity from the patient perspective. We have one of the largest real-world PRO datasets (prostate cancer) to our knowledge, incorporating 16,896 surveys on QoL and patient-reported toxicity from 1,948 prostate cancer patients treated with radiation. Second, we will use dosiomics, a recently developed methodology to create very high-resolution, three-dimensional maps of radiation dosage. Dosiomics can be used to identify radiation dosage with much greater precision than before, including dosage to small areas (such as the levator ani muscle, which is important for bowel continence). Third, we will use radiomics, or imaging of novel tumor features such as shape and texture that may affect radiation outcomes. Fourth, we will use advanced deep learning techniques developed by our research group to predict clinician-rated toxicity and PROs from the above data with much higher accuracy than previous machine learning approaches. To enhance the real-world impact of our research, we will externally validate our prediction algorithms using a detailed, existing retrospective dataset from 794 prostate cancer patients treated with radiation in the VA. Across both datasets, we have 2,742 patients, of which 526 patients are Black and approximately 274 are Hispanic. Study aims are as follows: Aim 1. To develop deep learning models incorporating dosiomics and radiomics for actuarial multi-endpoint prediction of clinician-rated toxicities among prostate cancer patients treated with radiation. Aim 2. To develop deep learning models incorporating dosiomics and radiomics for actual multi-endpoint prediction of PROs among prostate cancer patients treated with radiation. Aim 3. To validate deep learning models for actuarial multi-endpoint prediction of clinician-rated toxicities and PROs among prostate cancer patients treated with radiation in the VA. Our research team is uniquely qualified to conduct this study. Dr. Jim (Initiating PI) has extensive expertise in large-scale collection, analysis, and interpretation of PROs in cancer patients. Dr. El Naqa (Partnering PI) is at the forefront of deep learning architectures in complex, multi-omic datasets in cancer. He also has expertise in radiation oncology, including dosiomics and radiomics. Drs. Dicker and Johnstone (Co-Is) are national leaders in radiation oncology for prostate cancer, with the influence to translate findings locally and nationally to improve care. Dr. Katsoulakis (Other Significant Contributor) is a radiation oncologist with expertise in prostate cancer and a national leader in quality of care within the VA system. The entire team is particularly committed to survivorship in men with prostate cancer, with numerous individual and collaborative efforts to advance PROs and data science to improve outcomes and quality of care in this patient population. Upon completion of this study, we will have fulfilled our short-term objective to identify the risk of short- and long-term toxicity and detriments in QoL based on precise spatial evaluation of radiation dosage. This knowledge can immediately impact
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 28, 2022
- Source ID
- W81XWH2210276
Entities
People
- Heather Jim
Organizations
- H. Lee Moffitt Cancer Center & Research Institute
- United States Army