From Overload to Insight: Why AI Is Becoming an Essential Partner in Oncology

If the image of a robot comes to mind when prompted to consider what artificial intelligence (AI) in medicine looks like, you’re not the only one, said Brian P. Mulherin, MD, who discussed the use of AI and quality improvement in a presentation delivered during the inaugural MiBA Community Summit.1 However, taking a step back, it’s important to consider the necessity and limitations of AI to process and aid clinicians in discerning an exponential growth of medical knowledge (Figure).

The average turnover of new medical knowledge is 73 days, Mulherin said, who questioned how many individuals would feel comfortable prescribing imlunestrant (Inluriyo) for adults with estrogen receptor–positive, HER2-negative, ESR1-mutated advanced or metastatic breast cancer, after receiving FDA approval just 3 days ago.2 “There’s an increasing need to help somebody, at scale, organize all this and deliver these insights to the clinician at the point of care,” Mulherin said. Moreover, with an estimated 30,000 global clinical trials ongoing in oncology as of 2024 and an approximated 50 new drug approvals per year,3,4 there is a clear need to establish better decision support pathways, especially for community oncologists who remain “uniquely vulnerable,” Mulherin said.

Mulherin is the medical director of the American Oncology Network and a hematologist oncologist at Hematology Oncology of Indiana in Indianapolis.

What Is the Current Ability of LLMs to Summarize Large Amounts of Text?

Although one can make the argument that the National Comprehensive Cancer Network (NCCN) Guidelines can serve as the primary source for consolidated guidance on the proper treatment pathways for patients with any malignant tumor, Mulherin noted that 88 separate guidelines currently exist and multiple updates per guideline can be made annually.1 They’re also “voluminous, verbose, and seemingly never-ending,” he said.

But can you input a guideline into a large language model (LLM) like ChatGPT or Llama and expect it to churn out succinct and accurate guidance? Not quite, Mulherin said, citing a study published in 2025 that evaluated the ability of ChatGPT-4.0 and Llama 2 to answer 3 complex clinical questions using the NCCN Clinical Practice Guidelines for the management of thyroid carcinoma.5 The answers were scored on a Likert scale: 5) Correct; 4) Correct, with missing information requiring clarification; 3) Correct, but unable to complete answer; 2) Partially incorrect; and 1) Absolutely incorrect. The authors concluded that ChatGPT-4.0 and Llama 2 demonstrated a limited but substantial capacity to assist with complex clinical decision-making relating to the management of thyroid carcinoma.

Other generative AI tools on the market include Gemini, Claude, and Perplexity.1 “These are all general LLMs, and these are trying to be trained on the entire corpus of human knowledge; that’s why they need access to all our data,” Mulherin explained. Other tools like OpenEvidence are domain restricted, and while it does pull from reputable sources like journal publications, abstracts, and conference proceedings, it does not have the ability to pull from the NCCN guidelines, highlighting a current gap in utility.

Can AI Help Me Understand What Biomarkers I Need to Test?

The NCCN guidelines recommend high-quality next-generation sequencing (NGS) for patients with advanced non–small cell lung cancer (NSCLC) for over 14 biomarkers.6 “How can AI help? AI can, at scale, look at all the patients within the organization, and find out who is getting tested and who is not getting tested, and then if you have an actual biomarker, remind the physician that this person has an actual biomarker to act on,” Mulherin explained.

What does that look like in practice? It could include an automated comprehensive genomic profiling notification to physicians, alerting them to test their patients with advanced NSCLC, or a call, email, or text dispatched through the electronic medical record.1 Such an approach was implemented at Hematology Oncology of Indiana, and what was found was striking. Seven months prior to the launch of targeted education around the need for testing in patients with stage III and IV NSCLC, 87.34% had undergone NGS. Seven months after the program’s launch, 99.41% of patients had NGS. “If you think about this at a population level, how many people are you missing out on?” Mulherin underscored.

“AI can prompt physicians: ‘Your patient has an EGFR mutation but isn’t on osimertinib [Tagrisso]—why not?’ It’s like a digital safety net,” Mulherin said in an interview with OncLive® preceding his presentation at the summit.

Although AI is already deeply entrenched in many industries, its current iterations in medicine are still a work in progress, Mulherin explained, noting that even with substantial improvements in the technology itself, AI will work in conjunction with physicians to improve the quality of care, ensure equitable access, and provide data for future research and development.

Disclosures: Mulherin disclosed employment, leadership, and ownership for American Oncology Network; honoraria from Daiichi Sankyo/Lilly, Caris Life Sciences, Jannsen Oncology, Kite/Gilead, Swedish Orphan Biovitrum, Incyte, TG Therapeutics, CTI BioPharma Corp, Alexion Pharmaceuticals, Apellis Pharmaceuticals, and Novartis; consulting or advisory roles for Apellis Pharmaceuticals, Novartis, Vertex, Pharmacosmos, BeiGene, and Meaningful Insight Biotech Analytics; speakers’ bureau fees from Apellis Pharmaceuticals; research funding from Apellis Pharmaceuticals, Pharmacyclics/Janssen, and Novartis (inst); and travel, accommodations, and expenses fees from Incyte, Apellis Pharmaceuticals, and Kite.

References

1. Mulherin B. AI and quality improvement. Presented at: MiBA Community Summit; September 27-28, 2025; Scottsdale, Arizona.
2. FDA approves imlunestrant for ER-positive, HER2-negative, ESR1-mutated advanced or metastatic breast cancer. FDA. September 25, 2025. Accessed September 28, 2025. https://www.fda.gov/drugs/resources-information-approved-drugs/fda-approves-imlunestrant-er-positive-her2-negative-esr1-mutated-advanced-or-metastatic-breast
3. Izarn F, Henry J, Besle S, et al. Globalization of clinical trials in oncology: a worldwide quantitative analysis. ESMO Open. 2025;10(1):104086. doi:10.1016/j.esmoop.2024.104086
4. Novel drug approvals for 2024. FDA. July 14, 2025. Accessed September 28, 2025. https://www.fda.gov/drugs/novel-drug-approvals-fda/novel-drug-approvals-2024
5. Pandya S, Bresler T, Wilson T, Htway Z, Fujita M, et al. Decoding the NCCN guidelines with AI: a comparative evaluation of ChatGPT-4.0 and Llama 2 in the management of thyroid carcinoma. Am Surg. 2025;91(1):94-98. doi:10.1177/00031348241269430
6. NCCN. Clinical Practice Guidelines in Oncology. Non-small cell lung cancer, version 8.2025. Accessed September 28, 2025. https://www.nccn.org/professionals/physician_gls/pdf/nscl.pdf