Integration of AI in Oncology: Implications for Research, Clinical Outcomes

Artificial intelligence (AI) technology is integrated into the services and products we use every day in our personal and professional lives, although we do not think about it very often. Many of us may not be aware of the AI-fueled personal assistants and tools designed to improve our efficiency and productivity that are readily available, linked to or embedded within computer software programs we use routinely. However, as we continue to learn about applications for AI in oncology, we can better understand how these approaches solve problems and change how treatments are being developed and made available to patients.

Machine learning, the mathematical and statistical processes that computers use to improve their performance, and deep learning, the multilayered artificial neuronal network operations that facilitate unsupervised learning, are subfields of AI that have been employed for decades to mimic and surpass the human brain’s ability to learn and solve problems. Oncology is particularly well suited for these applications because of the complex and data-intensive processes required for cancer risk, early detection, diagnosis, prognosis, treatment determinations, predictions for treatment response, complications, survival, and disease recurrence.

As of October 7, 2022, the FDA had approved 521 medical devices using AI/ machine learning, with the top applications in radiology, cardiovascular medicine, and hematology. The agency also released a new guidance stating that certain AI-powered clinical decision support tools should be regulated as medical devices and that software functions must meet specific criteria to be considered nondevice clinical decision support. Examples could include AI and machine learning in software designed to predict a patient’s clinical deterioration, risk scores or probability of a disease or condition, or the potential to identify patients with opioid addiction. The number of medical devices using AI/machine learning reviewed and cleared by the FDA has surged since 2017.

In oncology, machine-learning strategies have facilitated drug discovery, optimized combination therapy design and delivery, and enabled more precise classification of disease states. The implementation of AI throughout the spectrum of drug discovery, development, and administration provides the potential to develop better targeted therapies and optimize and sustain drug selection and dosing regimens. Molecular profiling companies are combining molecular and clinical outcomes data to identify new strategies to improve precision medicine. Pharmaceutical companies, health care systems, health technology companies, and dozens of start-up companies are using AI algorithms to fight cancer. They are partnering across health systems and various countries to use electronic health records; imaging; and genomic, clinical, and other data to improve cancer detection, risk prediction, diagnoses, and patient clinical outcomes.

Available publications demonstrate how machinelearning approaches can help identify patients at risk for early and late adverse effects (AEs) associated with anticancer therapies, clinical biomarkers for specific AEs correlated with chemotherapeutic agents, complications related to underlying malignancies, and clinical prediction strategies for specific toxicities. AI-driven, patient-directed risk assessment tools can help motivate patients to adhere to treatments and seek medical care.

Furthermore, findings from clinical trials support the use of AI algorithms to enhance subject accrual and improve survival outcomes. Ongoing trials seek to evaluate AI-based precision oncology clinical matching tools and other AI platforms to facilitate clinical trial enrollment, develop predictive models, and predict quality of life and other outcomes associated with cancer treatments.

Limitations from randomized clinical trials of machine-learning interventions include potential risk of bias due to lack of study participants from underrepresented racial and ethnic minority groups. Therefore, broad applications for AI algorithms require diverse and inclusive data sets.

Additionally, the black box characteristic of AI technologies due to the overlapping decision trees hides within convoluted layers of data interactions. Unlike linear algorithmic approaches that can be represented graphically, the details of how data are analyzed to make decisions are less amenable to graphic presentation because of the nature of machine-learning operations.

Multiple AI-powered cancer clinical trial matching services and applications target patients for clinical trial recruitment, empowering them with the elimination of the oncologist as a gateway to clinical trials and removing additional barriers to clinical trial participation. The recruitment programs are available online and on mobile phone applications where patients can self-report information about their cancer history and submit clinical and test results. These approaches have the potential to expand the numbers of clinical participants, especially from a diverse population inclusive of underrepresented racial and ethnic minority individuals. Furthermore, pharmacists who are aware of these resources can better assist patients who are seeking clinical trials that may be outside their institution or region, especially if their tumor has biomarkers that could be incorporated into a trial query. Oncology pharmacists will likely interface with AI or machine-learning algorithms (if they do not already) to match patients to clinical trials; identify, select, and/or modify optimal treatments; or inform their approach to monitoring patients for treatmentassociated toxicities.

About the Editor

Lisa E. Davis, PharmD, FCCP, BCPS, BCOP, is the editor in chief of Pharmacy Times Oncology Edition. Davis holds positions as a clinical pharmacist in early-phase clinical trial and breast cancer programs at the University of Arizona (Arizona) Cancer Center and a clinical professor of pharmacy practice and science at the Arizona R. Ken Coit College of Pharmacy. Davis also sits on th Hematology/Oncology Pharmacy Association Board of Directors and is a member of the Cancer Prevention and Control Program and scientific review committee at the Arizona Cancer Center.