Key Takeaways

• ASCO 2026 highlighted that AI is rapidly moving from concept to clinical implementation, with emerging tools that can help oncology APPs streamline documentation, enhance decision support, improve symptom monitoring, and expand access to clinical trials.
• For oncology APPs, the greatest near-term value of AI lies in reducing administrative burden and improving workflow efficiency, enabling more time for patient education, care coordination, and direct patient care without replacing clinical judgment.
• As AI adoption accelerates in oncology, APPs should lead the responsible integration of validated, institution-approved tools while maintaining human oversight, verifying AI-generated recommendations, and ensuring equitable, patient-centered care.

Following the 2026 American Society of Clinical Oncology (ASCO) Annual Meeting, Jasmin Hundal, MD, DipABLM, MS, MPH, discussed how artificial intelligence (AI) is becoming an increasingly practical tool for oncology advanced practice providers (APPs). In this video interview, she reviews current and emerging applications—including ambient documentation, electronic health record decision support, remote symptom monitoring, AI-assisted imaging and pathology, patient education, and clinical trial prescreening—and emphasizes that the most immediate impact for APPs will be reducing administrative burden while improving efficiency and allowing more time for patient-centered care. Her presentation underscores a key message from ASCO 2026: as AI becomes more deeply integrated into oncology practice, APPs will play a central role in its safe, effective, and equitable implementation.

Transcript

Hello, everyone. My name is Jasmin Hundal. I’m a graduating fellow from the Cleveland Clinic and an incoming faculty member at Emory University as a breast medical oncologist.

AI has definitely become one of the biggest topics in medicine. AI is much broader than many people realize. It includes predictive AI, which estimates outcomes such as recurrence, toxicity, or treatment response; generative AI, which produces text and summaries; and multimodal AI, which integrates clinical data, imaging, pathology, genomics, and other information.

Specifically for APPs, AI is best understood as a set of tools that can reduce administrative work, organize complex information, identify concerning clinical changes, and support decision-making. It is not an independent decision-maker and will not replace clinicians. Rather, it augments the clinical team’s attention and judgment.

As clinicians and APPs, you already encounter AI through ambient documentation that converts patient conversations into draft notes. Studies have shown that it reduces administrative burden and clinician burnout. AI is already being used for inbox triaging, message drafting, visit summaries, and patient instructions.

Other examples include electronic health record alerts, risk scoring, and clinical decision support tools. AI is also used in assisted radiology and pathology interpretation, guideline navigation, and literature synthesis platforms such as OpenEvidence.

Other areas where clinicians may encounter AI include electronic patient-reported outcome systems, remote symptom monitoring, wearables, patient education, translation services, symptom-triage chatbots, and, perhaps most excitingly, clinical trial prescreening and matching. Many of these tools operate in the background, so clinicians may receive AI-generated information without realizing it.

The most promising AI applications are those that address existing clinical problems. First and foremost is reducing documentation work. AI can generate structured draft documentation and assist with inbox management. Studies have shown reductions in clinician burnout. For example, a study by Olson and colleagues conducted across six U.S. health systems over 30 days demonstrated a reduction in burnout from 51.9% to 38.8%. It also reduced clerical task load after hours and allowed clinicians to devote more attention to patients.

AI is also being explored for remote symptom monitoring, allowing symptoms to be captured between visits to facilitate earlier intervention. Much of this work is still under clinical investigation but is very promising.

AI-assisted imaging and pathology are already being used and are incorporated into NCCN Guidelines in certain settings, such as prostate cancer. The FDA has also approved some AI-assisted tools for breast cancer applications.

Another important area is patient-facing AI, which can simplify patient education, generate multilingual educational materials, and support initial symptom triage. Finally, clinical trial matching has already been implemented at some institutions and has shown promising results.

The largest near-term impact will likely come from ambient documentation, inbox support, and electronic health record-integrated guidance and decision support, allowing clinicians to synthesize information more efficiently.

Remote symptom monitoring and automated triage will also help support clinical decision-making. Other areas where AI is likely to have a major impact include patient education, multilingual communication for diverse patient populations, and clinical trial prescreening.

One of the greatest benefits of AI is reducing administrative work and allowing clinicians to spend more time with patients. Ambient AI converts conversations into draft histories, creates review of systems, assessment, and plan documentation, and allows clinicians to review, finalize, and sign the note. It can also prepare visit summaries, draft patient instructions, draft inbox messages, and extract relevant information from the electronic health record. Some institutions have already implemented many of these capabilities, while others are in earlier stages of adoption.

AI may also reduce the time required to search guidelines, screen patients for clinical trials, and review longitudinal symptom data. However, efficiency is not automatic. Poorly integrated tools can create alert fatigue, inaccurate documentation, and additional verification work. The quality of the output is critically important, and clinicians must continue to review AI-generated information carefully.

One of the biggest misconceptions is that AI will take over clinical work. APPs and clinicians remain essential, and AI should only improve workflow.

Another misconception is that AI is synonymous with chatbots. In reality, many AI tools are predictive models embedded within pathology, radiology, monitoring systems, or electronic health records.

AI can also sound highly confident and precise. Although many answers are accurate, AI can hallucinate information, omit important context, and produce estimates that appear more certain than the underlying evidence supports. It is essential to verify sources and ensure the information is accurate. It is also important to use only AI tools that have been approved within your institution.

When evaluating an AI tool, clinicians should ask several questions. What specific task is the tool designed to perform? Was it validated in patient populations and practice settings similar to your own? How well does it perform across different racial, demographic, and clinical subgroups? Was it trained on primary care patients, oncology patients, or another population? Can the outputs be verified against electronic health record data, clinical guidelines, or primary literature? Who is accountable if the tool produces an incorrect result, and what is the fallback process?

ASCO has also created a framework emphasizing transparency, informed stakeholders, fairness, accountability, oversight, privacy, and human-centered implementation.

Overall, AI should function as a second reader or drafting assistant, not as the final authority. Recommendations should always be interpreted alongside disease stage, biomarkers, prior treatments, comorbidities, toxicities, patient preferences, goals of care, and access considerations.

When clinical judgment and AI output disagree, that disagreement should trigger closer review. Clinicians should examine the inputs, verify the recommendations, and determine whether the model is missing important clinical context. Accountability for the final clinical decision always remains with the clinical team.

One emerging development is the ASCO AI-powered Guideline Assistant, which helps clinicians locate recommendations more efficiently. It is valuable for navigation but never replaces interpretation of the full guidelines.

The broader evolution of AI is moving toward multimodal models that combine imaging, digital pathology, genomics, spatial proteomics, and circulating tumor DNA. Potential applications include more individualized estimates of recurrence risk, treatment response, improved biomarker classification, and earlier detection of residual disease and treatment resistance.

Another concept under development is the digital twin, which creates a continuously updated computer representation of an individual patient. Although this is a compelling long-term vision, many of these applications remain investigational and require prospective, multisite validation before routine clinical use.

APPs do not need to become programmers to use AI effectively. Important skills include understanding the differences between predictive and generative AI; evaluating intended use, external validation, calibration, and model performance; providing complete, structured clinical context when using AI tools; verifying outputs against primary sources and established guidelines; recognizing hallucinations, automation bias, and performance limitations; protecting patient privacy by using only approved platforms; developing clear triage and escalation workflows; and explaining AI-supported care to patients in understandable language while being transparent about its use.

APPs are well positioned to lead AI implementation because of their central roles in patient education, care coordination, treatment planning, and communication across the oncology team.

One final message is that the safest and most valuable AI is not necessarily the most sophisticated model. It is a well-validated tool used for a clearly defined purpose, embedded within a reliable clinical workflow, with a human clinician accountable for the final decision.