New research has illustrated the strides being made to apply modern artificial intelligence (AI) computing methods to oncology, according to new findings to be presented at the European Society for Medical Oncology (ESMO) Congress 2023.
Background
Researchers have long investigated the potential of AI to transform cancer care and improve patient outcomes.
According to Amara’s Law, individuals tend to overestimate the short-term impacts of technology but underestimate its long-term effects. However, caution may be warranted in response to the enthusiasm for newer technologies such as AI, machine learning, and big data analytics—which may require a more gradual introduction than in other sectors. Examples of their application in clinical practice have so far been limited to the triage of biopsy images, mammograms, and lung computed tomography (CT) scans used to screen patients for tumors and to some areas of cancer research. Nonetheless, the implementation of these technologies into mainstream oncology research and practice has been far from uniform, signaling potential barriers that risk slowing its adoption and the benefits it could bring to cancer prevention, screening, and care pathways.
Researchers have argued that existing, well-defined guidelines on cancer screenings and diagnoses are not applied in the same way, even within Europe, for reasons that may be both economic and cultural. They noted that emerging digital solutions may exhibit the potential to intervene upstream and prioritize patients for screening based on their medical records.
“There are already AI-based platforms that allow [for] the analysis of data routinely collected in electronic health records and medical imaging units, and which could support prevention and screening programs by identifying individuals at risk of developing [cancer],” explained Raquel Perez-Lopez, PhD, a radiologist at the Vall d’Hebron Institute of Oncology in Barcelona, who was not involved in the recent study. Dr. Perez-Lopez stressed that these resources have remained underutilized, and he attributed this to the lack of an adequate legal framework for patient data to be used in this way.
Overview of the Potential of AI in Oncology
Less-tangible, but equally important, applications of modern computing methods are currently transforming certain areas of cancer research. In the field of cancer genetics, many of the mutations included in modern genomic reports used to match patients with targeted therapies were identified by AI-based methods. The technology compared the genetic profiles of hundreds of thousands of patients to make predictions about their role in the development of cancer. These tools have also recently begun to be used more broadly to analyze various types of data in real-world evidence studies—which are gaining traction as a means of generating evidence in settings such as rare cancers, where traditional randomized clinical trials are not feasible—or to bridge the gap between results achieved in clinical trials and real-world patient outcomes.
Recently, ESMO published its “Guidance for Reporting Oncology Real-World Evidence” (GROW), which were developed to guide scientific reporting in this field; to cover the subject of AI-based technologies; and to harmonize research practices in oncology by providing detailed recommendations for the steps necessary to test, validate, and report real-world data accurately and transparently. Among these recommendations were considerations related to the use of AI algorithms for data analysis in real-world evidence studies—an inclusion that is necessary to capture the relevant oncology-specific considerations and anticipate future developments.
Implementing Digital Oncology Into Practice
“In the near future, we could see AI tools transform data processing within hospital information systems and electronic health records by making it possible to structure physicians’ free-text notes and summarize vast quantities of information at the press of a button, which will greatly facilitate the extraction of real-world data from medical records to generate new research insights,” highlighted Rodrigo Dienstmann, MD, Director of Oncoclínicas Precision Medicine in São Paulo, Brazil, and Editor-in-Chief of ESMO Real World Data and Digital Oncology. “Adopting a standard method to assess AI technologies with the same degree of reliability with which we can evaluate medicines in clinical trials will be key to maximizing their benefits, while ensuring their adoption does not increase the risk of bias that could cause inequalities in patient care,” he added.
Dr. Dienstmann explained that the manuscript addressed a likely upcoming scenario in which the data used for research may no longer be collected and structured by a human expert but instead may be processed and summarized by a machine. The researchers emphasized that real-world research powered by advanced data analytics is becoming increasingly ubiquitous as a complement to clinical trials and is beginning to spread within the regulatory agencies that use it in the authorization process of new therapeutic agents. Therefore, the ability to accurately interpret this kind of evidence may be an essential skill for all oncology professionals in the future.
However, the researchers revealed that oncologists as a whole may not be prepared for the increased use of AI-based technologies in oncology, with educational needs that will increase proportionally with the entry of AI into clinical workflows. “There is a lot of apprehension about the impact AI will have on the profession once machines outperform physicians in a number of their traditional repetitive tasks. We need to train [physicians] to use these tools wisely and confidently based on a clear understanding of their value and limitations, so that machines and [physicians] together achieve better results for patients than either of them could on their own,” Dr. Dienstmann concluded.
