Information asymmetry has shaped the clinical relationship for generations. Interoperability and artificial intelligence are starting to change who has the information—and who gets to ask the next question.

For most of modern medicine, one fact has quietly structured the relationship between clinician and patient: the clinician held most of the usable information.

The patient knew her symptoms, her body, her experience, and what mattered to her. The clinician had the medical vocabulary, the scientific literature, the guidelines, the treatment algorithms, and—often—the medical record itself.

That information asymmetry was not inherently sinister. Medicine requires expertise. A person who has spent years learning physiology, pathology, diagnosis, and treatment should know things a patient does not.

But information asymmetry also created enormous power. When patients could not easily see the evidence behind a recommendation, they had limited ability to question whether a clinician was practicing according to that evidence. They often did not know what questions to ask, much less how to challenge the answers.

This essay is the first in a series exploring how that dynamic is shifting, and how better access to actionable, understandable information is reshaping the clinical encounter.

I had become fascinated by pharmacogenetics. I learned that we already had the ability to identify genetic differences that could materially affect how some people responded to certain medications. The science was not equally strong for every drug, and testing was sometimes marketed far more broadly than the evidence justified.

But for some drug-gene relationships, the evidence was striking. Clopidogrel (better known by the brand name Plavix) was one of them.

Clopidogrel is prescribed to reduce the risk of dangerous blood clots in people with cardiovascular disease, including patients who have undergone procedures such as coronary stenting. It is a prodrug, meaning the body must convert it into its active form before it can adequately inhibit platelets.

An enzyme called cytochrome P450 2C19 (CYP2C19) plays an important role in that activation.

Some people carry loss-of-function variants in the gene that produces CYP2C19. They do not activate clopidogrel as effectively. The result can be less active drug, less platelet inhibition, and a greater risk that the medication will fail to provide the intended protection.

The United States Food and Drug Administration (FDA) added a boxed warning to clopidogrel labeling in 2010 explaining that the drug has diminished effectiveness in patients who are CYP2C19 poor metabolizers. The labeling notes that tests are available to identify CYP2C19 genotype and directs clinicians to consider another platelet P2Y12 inhibitor in patients identified as poor metabolizers.

I remember learning this and wondering:

How is that information not part of every relevant prescribing decision?

A decade ago, in a different prescribing situation involving a loved one who was a pre-teen, I asked a psychiatrist to order pharmacogenetic testing before choosing a medication.

Here’s what she said: “I prefer to do trial and error myself.” I pushed, and she eventually agreed to the testing. I’m glad she did. The medication she was leaning towards had a higher risk of side effects and ineffectiveness with all else being equal. She chose the other option once she had the test results.

There are many appropriate answers a clinician could have given me other than “I prefer to do trial and error myself.” Here are a few! 👇

“The evidence isn't strong enough for this particular drug.”

“This test won't change what I recommend. Here’s why”

“There are other factors that outweigh the genetic result, like x, y, z.”

“The available studies do not apply well to this patient.”

“This test has limitations, and here is why I don't think it will improve this decision.”

Those are clinical arguments.

“I prefer trial and error” is not.

If scientifically established information could materially affect the safety or effectiveness of a treatment, knowingly dismissing that information should require more than personal preference. Yet medicine's information asymmetry has historically made that distinction difficult for patients to enforce.

How many would know that a test existed?

How many could find information about the relevant clinical guideline?

How many could read it and understand whether it applied to their situation?

How many would feel comfortable challenging the physician sitting across from them?

The National Academies has explicitly recognized this problem in diagnosis. Its work describes patients and families as essential members of the diagnostic team while acknowledging that patients may hesitate to assert themselves because they fear being perceived as difficult. Health literacy, unfamiliar medical terminology, culture, language, and difficulty navigating the healthcare system can all limit meaningful participation.

Expertise has always mattered, but expertise combined with control of the information creates a very different kind of authority.

Artificial intelligence (AI) did not suddenly arrive in a perfectly connected healthcare system.

For decades, valuable clinical information has been stranded in different hospitals, physician practices, pharmacies, laboratories, insurers, and patient portals. A clinician cannot incorporate information into a decision if she cannot see it.

Anyone who works in healthcare knows that this problem is not wholly solved and that much fragmentation remains. But the infrastructure is changing.

Through the 21st Century Cures Act’s mandates on standardized APIs and information blocking, CMS rules expanding payer-to-payer and patient-access data exchange, and the nationwide framework of TEFCA, data is beginning to move.

Interoperability is steadily weakening the barrier between organizations, while generative AI is weakening the barrier between specialized medical language and the patient. Put them together, and both parties can finally use modern tools to make sense of the same information.

These developments rarely generate the excitement of a new artificial intelligence model, but they are just as important.

When a patient downloads their electronic discharge summary or opens a health app that automatically summarizes their new medications, the workflow changes.

They aren't starting from a blank page trying to invent a medical question. They are reading a summary that lists a newly prescribed drug, a diagnosis, and an AI-generated breakdown of the care plan.

Prompted by that synthesis, they can ask the next logical question:

“Why was Plavix chosen for my stent, what other options were considered, and what factors went into that decision?”

We have talked about patient engagement in healthcare for years.

Much of that conversation has focused on adherence: helping patients understand what clinicians want them to do and encouraging them to do it. But that is a very limited view of what an engaged patient can contribute.

The Agency for Healthcare Research and Quality (AHRQ) describes patient and family engagement as including patients and families as active members of the healthcare team. Its diagnostic-safety work specifically encourages patients and families to contribute information that clinicians may not otherwise have and to participate in identifying possible diagnostic breakdowns.

Artificial intelligence adds another capability. Patients can increasingly bring questions generated from evidence into the encounter.

A patient may notice that a laboratory result changed.

A caregiver may ask whether two medications interact.

A family member may ask what diagnoses have not yet been considered.

When a patient can independently find the relevant evidence, the information advantage shifts. Sometimes, they spot the question everyone else missed, making it also a patient-safety intervention.

Research on diagnostic safety already tells us why that matters. The Agency for Healthcare Research and Quality notes that communication breakdowns during patient-clinician encounters are an important contributor to diagnostic errors and has developed tools specifically designed to improve information exchange between clinicians, patients, and families.

There is an understandable response from clinicians whenever this argument comes up:

Artificial intelligence can be wrong.

It can misinterpret information. It can generate inaccurate answers. It can fail to understand context. It can overstate uncertain evidence. That should be a consideration for everyone working on healthcare artificial intelligence. But human clinicians are also vulnerable to error.

They have incomplete information. They get tired. They rely on memory. They experience cognitive bias. They cannot possibly remain current on every paper, guideline, drug interaction, genetic association, diagnostic alternative, and treatment pathway relevant to every patient they see.

A 2024 randomized clinical trial in JAMA Network Open illustrates how early we still are in learning to do that. Physicians given access to a large language model did not significantly outperform physicians using conventional resources on the study's measure of diagnostic reasoning. In an exploratory comparison, however, the large language model operating alone scored higher than the physician groups.

There will be times when the guideline does not fit the patient. There will be times when a decision-support tool is wrong. There will be competing risks. There will be incomplete evidence. There will be human factors that no model adequately captures.

That is precisely why we need expert clinicians. But expertise should be able to explain itself.

A clinician who appropriately departs from an evidence-based recommendation should be able to explain the reason.

A clinician who rejects the output of an artificial intelligence system should be able to identify what the system missed.

A clinician who chooses not to use a validated decision-support tool when it could materially improve a high-stakes decision should eventually expect someone to ask why. That “someone” may be a medical director, a quality committee leader, a malpractice insurer, and the patient themself.

Interoperability frees the data, decision support makes it actionable, and AI makes it understandable. When those forces collide at the point of care, unchallengeable clinical authority ends.

The physician who told me, “I prefer trial and error,” a decade ago gave an answer patients and their families will no longer tolerate.

Healthcare has spent years debating whether clinicians will adopt new technology. The real question is changing: What happens when not adopting it becomes indefensible?

This is the first essay in a series looking at how generative AI and data interoperability are changing the dynamics of the clinical encounter.

Patients have been bringing printouts and web searches into exam rooms for years. But asking a search engine a static question is entirely different from using generative tools to translate a 40-page health record, query clinical guidelines in plain language, and model follow-up questions in real time.

Over the coming weeks, I’ll be exploring what that shift means for diagnosis, clinical decision support, and professional accountability—and how higher-resolution patient expectations will ultimately redefine the standard of care.

U.S. Food and Drug Administration. Plavix (clopidogrel bisulfate) Prescribing Information, March 2010.
This FDA-approved label documents the addition of the boxed warning in March 2010 stating that clopidogrel has diminished effectiveness in CYP2C19 poor metabolizers, that genetic tests are available, and that alternative treatment strategies should be considered in identified poor metabolizers.
FDA: 2010 Plavix prescribing information

DailyMed. Current clopidogrel prescribing information.
Current labeling retains the boxed warning that clopidogrel's effectiveness depends on conversion to an active metabolite, principally through CYP2C19; tests can identify poor metabolizers; and another P2Y12 inhibitor should be considered in identified poor metabolizers.
DailyMed: Current clopidogrel prescribing information

U.S. Food and Drug Administration. Table of Pharmacogenetic Associations.
The FDA places clopidogrel and CYP2C19 among drug-gene relationships for which evidence supports therapeutic-management recommendations. It identifies intermediate and poor metabolizers as having lower active-metabolite concentrations and antiplatelet response and potentially greater cardiovascular risk.
FDA: Table of Pharmacogenetic Associations

Lee CR, Luzum JA, Sangkuhl K, et al. Clinical Pharmacogenetics Implementation Consortium Guideline for CYP2C19 Genotype and Clopidogrel Therapy: 2022 Update. Clinical Pharmacology & Therapeutics.
The Clinical Pharmacogenetics Implementation Consortium (CPIC) guideline provides genotype-based prescribing recommendations, including recommendations to avoid standard-dose clopidogrel when possible in CYP2C19 intermediate and poor metabolizers in acute coronary syndrome and percutaneous coronary intervention settings.
CPIC: CYP2C19 and clopidogrel guideline

Pereira NL, Cresci S, Angiolillo DJ, et al. CYP2C19 Genetic Testing for Oral P2Y12 Inhibitor Therapy: A Scientific Statement From the American Heart Association. Circulation. 2024;150:e129-e150.
The American Heart Association statement concludes that the totality of evidence supports CYP2C19 testing before oral P2Y12 inhibitor prescribing in patients with acute coronary syndrome or percutaneous coronary intervention. It also discusses implementation needs including rapid results and electronic health record integration.
Circulation: American Heart Association scientific statement

Rao SV, O'Donoghue ML, Ruel M, et al. 2025 ACC/AHA/ACEP/NAEMSP/SCAI Guideline for the Management of Patients With Acute Coronary Syndromes. Journal of the American College of Cardiology. 2025;85(22):2135-2237.
The current multisociety acute coronary syndrome guideline provides contemporary recommendations for antiplatelet therapy but does not make a recommendation regarding CYP2C19 genotyping.
JACC: 2025 Acute Coronary Syndromes Guideline

Cavallari LH, Lee CR, Beitelshees AL, et al. 2025 Acute Coronary Syndrome Guideline: Missing the Boat on CYP2C19 Genotyping. Journal of the American Heart Association. Published May 6, 2026.
This viewpoint specifically addresses the absence of a CYP2C19 genotyping recommendation in the 2025 guideline and reviews the evidence supporting genotype-guided selection of P2Y12 inhibitors.
Journal of the American Heart Association: Missing the Boat on CYP2C19 Genotyping

Office of the National Coordinator for Health Information Technology. ONC's Cures Act Final Rule.
The rule advances access, exchange, and use of electronic health information, establishes standardized application programming interface requirements, expands patient access to electronic information, and implements information-blocking provisions of the 21st Century Cures Act.
HealthIT.gov: Cures Act Final Rule

Office of the National Coordinator for Health Information Technology. Information Blocking.
This federal resource explains the prohibition on practices likely to interfere with the access, exchange, or use of electronic health information and describes current enforcement and provider requirements.
HealthIT.gov: Information Blocking

Office of the National Coordinator for Health Information Technology. Trusted Exchange Framework and Common Agreement.
The Trusted Exchange Framework and Common Agreement (TEFCA) establishes a nationwide network-of-networks intended to allow providers, payers, patients, and public-health organizations to securely exchange health information across organizational and proprietary boundaries.
HealthIT.gov: TEFCA

Office of the National Coordinator for Health Information Technology. Data Liquidity, Affordability, and Access: The History & Growth of TEFCA. 2026.
This recent federal update documents the rapid expansion of nationwide exchange, reporting growth from approximately 10 million documents exchanged through TEFCA before 2025 to 464 million by the end of 2025.
HealthIT.gov: History and growth of TEFCA

Centers for Medicare & Medicaid Services. CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F).
The rule expands interoperable data exchange through Patient Access, Provider Access, Payer-to-Payer, and Prior Authorization application programming interfaces, with major application programming interface requirements taking effect beginning in 2027.
CMS: Interoperability and Prior Authorization Final Rule

National Academies of Sciences, Engineering, and Medicine. Improving Diagnosis in Health Care. 2015.
The National Academies concluded that patients and their loved ones should be central members of the diagnostic team and identified information, communication, health-literacy, cultural, and power barriers that can inhibit meaningful patient participation.
National Academies: Improving Diagnosis in Health Care

Agency for Healthcare Research and Quality. Patient and Family Engagement.
The Agency for Healthcare Research and Quality (AHRQ) defines patient and family engagement as including patients and families as active members of the healthcare team and collaborative partners with clinicians and healthcare organizations.
AHRQ: Patient and Family Engagement

Agency for Healthcare Research and Quality. Toolkit for Engaging Patients to Improve Diagnostic Safety.
The toolkit is designed to help patients, families, and clinicians work together to improve diagnostic safety through better communication and information sharing during clinical encounters.
AHRQ: Toolkit for Engaging Patients to Improve Diagnostic Safety

Agency for Healthcare Research and Quality. Answer the Call to Engage Patients and Families in the Diagnostic Process.
This AHRQ-funded work tested approaches that allow patients and families to contribute information to the diagnostic process. Research using the OurDX tool found that patient and family contributions can uncover diagnostic concerns and potential blind spots that otherwise may not be visible to clinicians.
AHRQ: Patient and family contributions to diagnostic safety

Montero A, Montalvo J III, Kearney A, et al. KFF Tracking Poll on Health Information and Trust: Use of AI for Health Information and Advice. March 25, 2026.
KFF found that 32 percent of U.S. adults had used artificial intelligence tools or chatbots for health information or advice in the previous year.
KFF: Use of AI for Health Information and Advice

Goh E, Gallo R, Hom J, et al. Large Language Model Influence on Diagnostic Reasoning: A Randomized Clinical Trial. JAMA Network Open. 2024;7(10):e2440969.
The randomized trial examined whether access to a large language model improved physician diagnostic reasoning and found that providing access to the tool did not, by itself, significantly improve physicians' diagnostic-reasoning scores.
JAMA Network Open: Large Language Model Influence on Diagnostic Reasoning

Olson KD, Meeker D, Troup M, et al. Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout. JAMA Network Open. 2025;8(10):e2534976.
In a multicenter quality-improvement study of 263 ambulatory clinicians, ambient artificial intelligence scribe use was associated with reduced burnout and documentation burden and improved clinicians' perceived ability to give patients undivided attention.
JAMA Network Open: Ambient AI Scribes and Professional Burnout

Duggan MJ, Gervase J, Schoenbaum A, et al. Clinician Experiences With Ambient Scribe Technology to Assist With Documentation Burden and Efficiency. JAMA Network Open. 2025;8(2):e2460637.
This study found that ambient scribe use was associated with less time spent documenting, less after-hours work, lower perceived documentation burden, and a greater sense of engagement with patients.
JAMA Network Open: Clinician Experiences With Ambient Scribe Technology

Mello MM, Guha N. Understanding Liability Risk from Using Health Care Artificial Intelligence Tools. New England Journal of Medicine. 2024;390:271-278.
Mello and Guha examine emerging malpractice and liability questions associated with clinical artificial intelligence, including how conventional negligence principles may apply as artificial intelligence becomes embedded in care.
New England Journal of Medicine: Understanding Liability Risk from Using Health Care AI Tools

*Disclaimer: All opinions and ideas expressed in this article are solely mine and none represent a recommendation or should be viewed as advisement of any kind to anyone to do anything.*