Cardiovascular Certification Must Measure the Competencies AI Cannot Replace

The rejection of the American Board of Cardiovascular Medicine (ABCVM) application by the American Board of Medical Specialties (ABMS) in February 2025 should not end the debate about how cardiologists are certified, but it should intensify it.

Almost a decade ago, I discussed the disruptive potential of AI and the future of physicians in the era of precision medicine.¹ Today, the fundamental question is no longer simply whether cardiovascular medicine needs certification. It is whether the current model of certification measures the competencies that matter most to contemporary cardiovascular practice.

Cardiology has changed profoundly since cardiovascular medicine became part of the American Board of Internal Medicine (ABIM) structure. The field has become increasingly specialized, procedural, imaging intensive, technology driven, overwhelmed by rapidly emerging clinical trials, and multidisciplinary. Interventional cardiology, electrophysiology, advanced heart failure, structural heart disease, cardiovascular imaging, cardio oncology, critical care cardiology, preventive cardiology, and cardio-obstetrics represent distinct areas of expertise with rapidly evolving knowledge and technical requirements.²⁻⁴

The profession itself has recognized this evolution. In 2023, cardiovascular leaders argued that certification should move toward continuing clinical competence, knowledge gap identification, and quality improvement rather than an emphasis on periodic testing. Furthermore, multiple studies have demonstrated that AI systems are increasingly capable of performing at or above physician level on examinations of medical knowledge.⁵˒⁶ In 2024, ACC leadership further articulated the rationale for an independent cardiovascular board capable of strengthening continuous competency and professional growth.

The ABCVM proposal therefore addressed a legitimate problem.

Certification Should Measure Competence, Not Just Knowledge

The traditional certification model was developed when medical knowledge was the principal scarce resource. A physician who could demonstrate mastery of a large body of knowledge through a standardized examination had demonstrated an important component of competence.

That remains necessary.

But it is no longer sufficient.

The emergence of large language models (LLMs) and increasingly capable clinical AI systems is changing the nature of medical expertise. AI can retrieve information, summarize medical records, generate differential diagnoses, recognize patterns in images and electrocardiograms, and increasingly assist with clinical decision making.

The critical skill is therefore shifting from simply knowing information to knowing how to use AI generated information safely and appropriately. This is similar to knowing how to use Google Search or PubMed effectively. The skill is not simply finding information, but knowing what to search for, how to evaluate the quality of the information, and how to determine whether it applies to the clinical question at hand.

In the coming decades, cardiologists will increasingly use AI assisted tools, including frontier models and specialized or vertical AI models such as CardioOracle, UpToDate, OpenEvidence, and tools that do not yet exist.⁷ These tools may outperform cardiologists on standardized examinations of medical knowledge, including the USMLE and ABIM board style examinations. There is little doubt that this trend will continue.

A cardiologist will therefore need to recognize when an AI generated interpretation of an echocardiogram, coronary angiogram, myocardial perfusion study, or CCTA is plausible, when it is uncertain, and when it is wrong. The clinician must understand the limitations and validation population of a machine learning risk model before incorporating its output into patient care.

Consider commonly used clinical risk tools such as CHA₂DS₂ VASc, H₂FPEF, GRACE, or HAS BLED. The important competency is not simply remembering the components of the score. It is understanding what the score can and cannot tell us about an individual patient and integrating that information into clinical judgment and shared decision making.

These competencies are difficult to measure with a static, closed book examination.

They are better evaluated longitudinally, in context, and closer to actual clinical practice.

The Next Generation of Certification

Testing medical knowledge alone is becoming outdated. The ABCVM should therefore not simply attempt to create another version of the existing examination.

It should build a different model.

A knowledge based examination in the AI era is like using a fax machine when email is available.

Certification could become a longitudinal portfolio of evidence demonstrating continued competence, along with the ability to appropriately utilize clinical AI tools. Knowledge assessments would remain an important component, but they should be integrated with practice improvement, clinical AI skills, procedural and imaging competency, continuing education, clinical performance, and other meaningful measures of professional development.

Emerging technology could make this model feasible.

AI could identify individual knowledge gaps and generate personalized educational pathways. Learning could be tailored to the cardiologist’s subspecialty, practice environment, and demonstrated areas of need. An electrophysiologist should not necessarily receive the same learning pathway as an interventional cardiologist, heart failure specialist, preventive cardiologist, critical care cardiologist, or cardiovascular imager.

This is not an argument for replacing physician judgment with AI.

It is the opposite.

The more powerful AI becomes, the more important it is to assess the physician’s ability to supervise, interrogate, and appropriately override it.

The certification system should evolve accordingly.

From Maintenance of Certification to Maintenance of Competence

This distinction is important.

The objective should not be to eliminate accountability or professional standards. Cardiologists should continue to demonstrate that they remain knowledgeable, competent, and capable of providing safe and high quality care.

But certification should increasingly focus on maintenance of competence and the ability to use AI tools appropriately, rather than simply maintenance of certification.

The difference is more than semantics.

A physician who completes meaningful quality improvement activities, demonstrates appropriate procedural performance, identifies and closes knowledge gaps, participates in relevant continuing education, and maintains competency in a rapidly changing specialty should receive meaningful credit for those activities.

Conversely, simply completing educational modules or periodically passing an examination should not, by itself, be considered sufficient evidence of continued clinical competence.

The goal should be to create a system in which certification is useful to the physician, meaningful to the profession, leverages emerging technologies, and remains credible to the public.

On Chain Figure demonstrates a proposed portfolio that uses AI to map individual knowledge gaps and generate subspecialty specific learning pathways, then integrates targeted knowledge assessment, clinical AI supervision, technical performance, and practice improvement.

The Case for Cardiovascular Leadership

An independent cardiovascular board would not solve every problem. Developing valid measures of clinical competence is difficult. Procedural specialties require different assessment strategies than nonprocedural fields. AI introduces new challenges involving bias, validation, transparency, automation bias, and accountability.

These challenges, however, are arguments for innovation, not reasons to preserve the status quo.

The cardiovascular community has the expertise, professional societies, clinical data infrastructure, and educational ecosystem necessary to develop a modern certification framework. The ABCVM effort brought together the major cardiovascular organizations, the ACC, AHA, HRS, HFSA, and SCAI, to pursue such a model.

The rejection of the ABCVM application does not invalidate the underlying problem.

If anything, it makes the next step more important.

Cardiovascular medicine should continue to pursue an innovative certification model that is specialty specific, continuously updated, individualized, technology enabled, and grounded in demonstrated clinical competence.

The next generation of cardiologists should not be certified primarily on their ability to remember what was known yesterday or a few years ago.

They should be assessed on their ability to practice safely and intelligently in a world in which what is known and how it is accessed changes every day.

The ABCVM fight is not over. It should not be.

AI Disclosure: CK wrote and edited the article, including its ideas, analysis, and conclusions. An AI model was used solely for grammar and spelling edits. The central illustration was generated using AI based on CK’s ideas and refined through CK’s inputs.

References

1. Krittanawong C. Future Physicians in the Era of Precision Cardiovascular Medicine. Circulation. 2017;136(17):1572-1574. doi:10.1161/CIRCULATIONAHA.117.029676. PMID: 29061572.

2. Shapiro MD, Maron DJ, Morris PB, et al. Preventive Cardiology as a Subspecialty of Cardiovascular Medicine: JACC Council Perspectives. J Am Coll Cardiol. 2019;74(15):1926-1942. doi:10.1016/j.jacc.2019.08.1016. PMID: 31601373.

3. Braunwald E. Cardio obstetrics: a new specialty. Eur Heart J. 2024;45(18):1589-1592. doi:10.1093/eurheartj/ehae202. PMID: 38569057.

4. Sinha SS, Geller BJ, Katz JN, et al. Evolution of Critical Care Cardiology: An Update on Structure, Care Delivery, Training, and Research Paradigms: A Scientific Statement From the American Heart Association. Circulation. 2025;151(10). doi:10.1161/CIR.0000000000001300. PMID: 39945062; PMCID: PMC12063187.

5. Miranda J, Pereira-Silva R, Guichard J, et al. Artificial Intelligence Outperforms Physicians in General Medical Knowledge, Except in the Paediatrics Domain: A Cross-Sectional Study. Bioengineering (Basel). 2025;12(6):653. doi:10.3390/bioengineering12060653. PMID: 40564469; PMCID: PMC12190018.

6. Katz U, Cohen E, Shachar E, et al. GPT versus resident physicians: A benchmark based on official board scores. NEJM AI. 2024;1(5). doi:10.1056/AIdbp2300192.

7. Vishwanath K, Alyakin A, Ghosh M, et al. General-purpose large language models outperform specialized clinical AI tools on medical benchmarks. Nat Med. 2026;32(7):2405-2409. doi:10.1038/s41591-026-04431-5. PMID: 42286322; PMCID: PMC13375586.

8. Peri K, Eisenberg MJ. Commentary: Review of Mandatory Maintenance of Certification in the USA. Eval Health Prof. 2025;48(4):494-499. doi:10.1177/01632787241311360. PMID: 39749561; PMCID: PMC12518884.