Fall 2025 Issue

Precision Tool
By Selena Chavis
For The Record
Vol. 37 No. 4 P. 10

Experts weigh in on the promise of integrating AI-driven precision medicine in EHRs.

Two decades ago, EHRs formed the basis of the health IT movement, transforming data collection and providing a crucial alternative to paper-based patient records. Today, these infrastructures have evolved from repositories housing important information to conduits for making precision medicine a reality. According to Cillian Cheng, a PhD candidate with the department of medicine and therapeutics with Chinese University of Hong Kong, health care has witnessed the integration of precision medicine into EHRs as it has gone from a conceptual goal to an active area of technological and clinical development.

“This evolution is characterized by a strategic shift from a primarily genomics-focused approach to a more holistic, data-driven model that leverages advanced analytics,” he says.

AI is an important factor in this evolution. EHR integration of precision medicine is moving toward a dynamic, AI-enabled, clinically embedded framework aimed at delivering personalized, predictive, and equitable health care. John Mueller, director of commercial programs with Foundation Medicine, says that new data-driven, multiomic, AI-enabled ecosystems for precision medicine are reshaping drug development, clinical care, and patient engagement worldwide.

“In that time, EHRs have started partnering with precision medicine companies like Foundation Medicine to integrate their offerings into EHR workflows and streamline treatment decision making for patients,” he says. “These integrations include comprehensive genomic profiling ordering and results delivery, discrete biomarker data availability, and AI-driven clinical decision support.”

In fact, most of the heavy hitters in the EHR space—Oracle, Epic, Meditech, and Allscripts, to name a few—have made announcements in recent years about integrating precision medicine into clinician workflows. Earlier this year, Epic, in particular, expanded its application programming interface and interoperability frameworks to support genomic data exchange for more than 2,000 hospitals.

Cheng points out that AI and natural language processing have become essential for interpreting both structured and unstructured EHR data, enabling improved patient phenotyping, risk prediction, and personalized treatment planning. Speaking directly to the use of precision medicine in EHRs for oncology, he adds that greater emphasis is now placed on clinical actionability, with decision support tools and precision oncology models demonstrating how integrated molecular data can guide real-world therapy choices. The scope of data is also expanding to include social determinants of health and patient-generated digital data from wearables and apps.

Cheng recently authored a 2025 study in Molecular Cancer that synthesized the current landscape of AI in oncology. Titled “Artificial Intelligence in Cancer: Applications, Challenges, and Future Perspectives,” findings suggest that the most powerful applications “integrate multimodal data, including medical imaging (CT, MRI, histopathology), genomics, and EHRs, using sophisticated deep learning models such as convolutional neural networks and transformers.”

Foundation Medicine was one of the pioneers of bringing genomic data for cancer care into EHRs, such as Epic. According to Mueller, the primary use of the Epic integration is for health care providers to order high-quality biomarker tests and receive results seamlessly.

Current movements are providing momentum for more integration of precision medicine in EHRs, but Cheng notes that the challenges are many. “While numerous AI models have demonstrated radiologist- or pathologist-level performance in retrospective studies, their widespread clinical adoption faces significant barriers, including data privacy concerns, algorithmic bias, and the need for robust regulatory frameworks and prospective validation,” he says.

EHR Challenges
The promise of EHR-based precision medicine is obvious to most stakeholders, but there are significant hurdles that health care systems must overcome, such as interoperability, data standards, computational demands, and persistent equity gaps, especially the underrepresentation of diverse populations. “A major hurdle is the lack of representativeness in training datasets, which often overrepresent populations of European ancestry,” Cheng says. “This can lead to models that perform poorly and even exacerbate health disparities when applied to underrepresented groups.”

On the interoperability front, Mueller points out that the lack of standardized data formats and the absence of native supports within EHRs for genomic data continues to present challenges. “There are dozens of ways to ingest the genomic information we produce in our reports and deliver it in a format that’s easily ingestible, and you add more complexity when you layer an EHR on top of that,” he explains. “To address this, we work closely with our EHR partners to find a streamlined and easily digestible way to present our results within their systems.”

Mueller adds that establishing appropriate data privacy and consent management will be important to advancing the promise of precision medicine. “We all recognize that genomic information is highly sensitive, and, as a result, compliance with HIPAA, [the General Data Protection Regulation], and other regulations, though necessary, can add complexity to integration of precision medicine capabilities into EHRs,” he explains.

The same applies to regulatory pathways, Cheng says, pointing to the complexities of defining AI as a medical device. “Many AI models lack rigorous external validation on independent, multicenter cohorts and prospective studies, which is crucial for proving real-world clinical utility and gaining trust,” he says.

Finally, with the increasing amount of information becoming available within EHRs to support treatment decision making, Mueller says there is the added risk for decision fatigue for the health care providers who are already bandwidth constrained. Overcoming this challenge will require providing in-depth trainings and educational resources to simplify the onboarding process so health care teams can focus on providing the best care to patients, he says.

Precision Medicine’s Promise
Challenges aside, the promise of EHR-driven precision medicine is real. Cheng cites multiple compelling examples in the field of oncology from his 2025 study related to colorectal cancer, breast cancer, and prostate cancer. In colorectal cancer, AI systems for computer-aided detection during colonoscopy have received FDA clearance and, in randomized trials, have been shown to increase adenoma detection rates, potentially preventing interval cancers. Additionally, AI models such as Mirai can now predict a patient’s five-year risk of developing breast cancer directly from mammograms, enabling personalized screening strategies. For prostate cancer, CE-marked and FDA-approved AI systems such as Paige Prostate can detect cancer in biopsy specimens with high accuracy, serving as a valuable second read for pathologists that improves diagnostic accuracy and consistency.

The drug discovery realm of health care is another exciting opportunity for EHR-driven precision medicine. Cheng points to generative AI models that have identified a novel potent kinase inhibitor in just 21 days, a process that traditionally takes years.

A 2024 study conducted at the University of Florida and published in Frontiers in Pharmacology used HL7 standards to integrate pharmacogenomic data into their Epic “Genomic Module” to make it actionable. Titled “Integration of Pharmacogenetic Data in Epic Genomic Module Drives Clinical Decision Support Alerts,” the study had genomic indicators flag actionable genetic information directly in the EHR. These indicators feed into clinical decision support alerts so that when a clinician prescribes a drug, the system can warn about gene–drug interactions or recommend alternative dosing.

What’s notable about the study is that the approach moves away from having genetic info only in nonstructured lab reports, making pharmacogenics much more integrated and usable. Ultimately, the authors note that findings reveal real clinical decision support using genomic data—not just storage—helps clinicians make better prescribing decisions.

Mueller points to a 2022 study done at the University of Pennsylvania that demonstrated significantly streamlined delivery of genomic medicine due to integrating genomic test results into EHR workflows. Clinicians can now view actionable variants directly in the patient’s chart, eliminating delays caused by separate portals. This means patients receive targeted therapies sooner, improving outcomes in oncology care.

The use cases are numerous and point to real opportunities for efficiently integrating precision medicine into EHR workflows in real clinical settings. When genomic data is represented in structured formats, EHRs can provide clinical decision support so clinicians can act on genetic information. Yet, while process improvements are well documented, hard clinical outcome data is less mature or harder to demonstrate broadly, according to the research.

“We continuously seek feedback from our customers with EHR integrations, and we’ve heard many positive stories about the impact these integrations have had in their clinics,” Mueller says. “This includes everything from improved ease of ordering our tests and reading our results reports to quicker turnaround times and an improved tracking experience.”

Clinicians’ Future Role
Cheng believes that a crucial and often overlooked aspect of current movements with AI-driven precision medicine in EHRs is the human element and the future role of clinicians.

“The goal of AI in precision medicine is not to replace doctors but to augment their capabilities,” he emphasizes, noting that a powerful example is the emergence of AI-powered ambient clinical intelligence. “These systems act as digital scribes, understanding doctor-patient conversations and automatically generating clinical notes in real-time.”

Cheng adds that health care organizations using this technology report that physicians reclaim two to three hours per day, directly addressing burnout and allowing oncologists to focus on what matters most: complex decision making and empathetic patient interaction. As the industry continues to discuss the promise of AI, he notes that it must also “champion the irreplaceable value of human experience and judgment, particularly in providing compassionate care and understanding the nuanced, personal journey of a cancer patient. The most successful future health care model will be a synergistic partnership between human expertise and AI.”

— Selena Chavis is senior director of accounts with Insenna and a Florida-based
freelance writer.