A workshop group assessed the use of multidimensional data obtained from patients with cancer and the computational methods used to analyze the data.
A workshop review published in JAMA Oncology evaluated the best practices for developing and assessing the clinical utility of predictive computational methods in oncology.
“In many ways, the use of computers to diagnose disease and guide treatment has led to more questions than answers,” the authors wrote. “Unfortunately, the excitement to harness computers in the oncology clinic has led to several high-profile instances of premature or inappropriate use of computational predictive systems in cancer before they been adequately tested and validated.”
The National Cancer Policy Forum and the Board on Mathematical Sciences and Analytics at the National Academies of Sciences, Engineering, and Medicine hosted the workshop to assess the use of multidimensional data obtained from patients with cancer and the computational methods used to analyze the data.
Key characteristics for successful computational oncology were considered in 3 main thematic areas, including:
Overall, quality control was found to be essential across all stages. Additionally, the workshop group suggested that collecting a “standardized parsimonious” data set at every cancer diagnosis and restaging could help improve reliability and completeness of clinical data for precision oncology.
“A finalized list of cancer-relevant variables for collection has yet to be defined, to our knowledge; however, there are cancer centers with first-hand experience in establishing a minimum data set per patient, and further discussion on this topic will benefit from participation by tumor registrars and cancer registry staff,” the authors wrote.
However, the reliability of data used to develop and use computational algorithms is critically dependent on their completeness, quality, diversity, relevancy, timeliness, and accuracy, with potential biases or limitations being proactively managed.
Data completeness, according to the review, refers to key data elements such as information about cancer diagnosis, treatment, and outcomes, while data quality depends on whether appropriate variables have been measured in valid and reliable ways. Moreover, data diversity can be improved by collecting data from diverse groups, reducing the risk of creating invalid and biased algorithms.
“The inadequacy of data sets representative of patient population diversity can have significant consequences, such as computer algorithms incapable of recognizing melanomas on black individuals because the algorithms were trained predominantly to read lesions on individuals with lighter skin,” the authors wrote. “Documenting data reliability, not only in source data sets, but also in derived data sets used for final analyses and data set generation, is essential in the creation of computational methods for eventual clinical application.”
Currently, treating oncologists generally rely on pathologists and bioinformatics specialists to properly use computational methods when analyzing high-dimensional molecular cancer data. Even further, many clinical oncologists tend to be limited by time and training in vetting and interpreting computational outputs for their patients. For this reason, the workshop participants stressed the importance of utilizing a multidisciplinary team.
“A deep understanding of computational methods requires knowledge and experience in both molecular cancer diagnostics (e.g., next-generation sequencing) and computer programming (e.g., Python),” the authors wrote. “These skillsets in addition to clinical oncology training are rare, and it is impractical to expect all clinical oncology fellows to learn these computational skills for 21st-century oncology.”
“Instead, it is more feasible for clinical oncologists, pathologists, and bio- informatics engineers to work more closely together as a multidisciplinary team,” the authors continued. “A natural venue for this activity is tumor boards or precision oncology boards, which meet regularly, focus on the patient, and are designed to elicit communal feedback.”
In addition, the review highlighted the need for cancer clinics and hospitals to ensure the adequate support of computational oncology. This would include hiring staff and purchasing computer resources for assembling and structuring data, creating reports that allow oncologists to visualize and act on the data, and communicating securely with other computational systems.
“Precision oncology therapies, which target specific genetic changes in a patient’s cancer, are changing the nature of cancer treatment,” the authors concluded.
Panagiotou OA, HÃ¶gg LH, Hricak H, et al. Clinical Application of Computational Methods in Precision Oncology. JAMA Oncology. doi:10.1001/jamaoncol.2020.1247.