AI in Health Care: New Biomarkers, Drug Candidates, and Growing Adoption

The use of artificial intelligence in health care has grown rapidly, with more than 80% of U.S. physicians reported using AI in their work in 2025 — more than twice the percentage in 2023 — according to a survey by the American Medical Association. More than 70% of respondents said they think AI can help improve health care, mainly by making the system more efficient and reducing burnout. Meanwhile, new research demonstrates AI's ability to uncover hidden disease biomarkers and accelerate drug discovery.

Researchers at the University of California, Berkeley and collaborating institutions trained a deep learning model using more than 440,000 electrocardiograms linked to national death records in Sweden. The model discovered a previously unrecognized biomarker within routine ECGs that identified patients at exceptionally high risk for sudden cardiac death, many of whom would not have been identified using current clinical guidelines. In radiology, recent research shows AI systems can detect subtle bone metastases on CT scans with accuracy comparable to or exceeding experienced radiologists, and can improve radiologists' performance when used as an assistive tool.

In drug discovery, Mindbeam AI used a combination of generative AI, computational modeling, and virtual screening to design and evaluate 24 novel drug candidates targeting TRPV1, a receptor involved in pain signaling. Three lead compounds emerged with strong potential; one exhibited a favorable balance of predicted efficacy, bioavailability, and tolerability. The work aims to address the liver toxicity risk associated with chronic high dosing of acetaminophen, one of the most widely used over-the-counter pain relievers. "This is just the beginning of what's possible beyond acetaminophen," said the founder and CEO of Mindbeam AI.

AI is also being integrated into clinical trial infrastructure, where concerns about equity persist. A commentary in Nature notes that underrepresentation of racial and ethnic minorities, women, older adults, pregnant individuals, and socioeconomically disadvantaged populations in clinical trials remains a chronic problem. Documented harms include a widely deployed health-management algorithm that used healthcare spending as a proxy for need, systematically under-prioritising Black patients, and an evaluation of nine large language models across over 1.7 million outputs that found biased recommendations for cases labelled as Black, unhoused, or LGBTQIA+. The same analysis argues that appropriately governed AI can also reduce disparities.

The commercial market reflects the rapid adoption. According to a report by Grand View Research, the global AI in healthcare market was estimated at USD 36.7 billion in 2025, is anticipated to grow from USD 50.7 billion in 2026 to USD 505.6 billion by 2033, expanding at a compound annual growth rate of 38.90% during the forecast period.

Industry executives share the optimism. More than 80% of health executives say AI will be invaluable this year across different operations in the industry, according to Deloitte's 2026 U.S. Health Care Outlook. AI is already helping with routine documentation, analyzing patients' medical history, improving liquid biopsies for lung cancer or brain tumors, speeding up MRI diagnoses, reducing false alarms, and predicting disease risk. At the same time, lawmakers have raised concerns about patient privacy and the potential for AI-driven systems to carry or worsen biases based on race and gender; Senate hearings have examined AI integration in health care.

However, some early research suggests AI use can erode critical thinking, creativity, and skill development. In a 2025 experiment at MIT, people's brain activity was reduced when they wrote an essay with AI assistance. Another study analyzing 370,000 college-admission essays found that after the advent of generative AI chatbots, the range of original ideas shrunk. Medical educators are developing methods to integrate AI while preserving critical thinking, such as having students critique AI output or solve problems without laptops.

AI is also enabling personalized treatment at an individual level. OpenAI's CEO highlighted a case in which an individual used ChatGPT and other large language models to design a personalized mRNA vaccine for his dog diagnosed with cancer. The process involved genomic sequencing, identification of a c-KIT gene mutation, use of AlphaFold to investigate protein behavior, and discovery of neoantigens for vaccine design. The final treatment combined the custom vaccine with a tyrosine kinase inhibitor and a PD-1 inhibitor, with timing and interactions optimized using AI. The CEO noted that human specialists remained necessary at each stage.