Can AI-powered EBM change the U.S. health care system?
AI has the potential to change many parts of our global community. However, maybe none as far-reaching as its potential impact on health care systems worldwide. This article will focus on the possibilities and challenges of building an enhanced evidence-based medical model to improve the U.S. health care system. This change will undoubtedly require patients, providers and the health care industry to overcome many hurdles—the greatest of which is change. The health care system has historically been very slow to adapt to change.
What is EBM?
Evidence-based medicine (EBM) has been touted as the cornerstone of modern, high-quality health care. Defined by the rigorous, conscientious and judicious integration of the best available research evidence, clinical expertise and patient values, EBM serves as the critical mechanism for moving scientific discovery from research lab to clinical care. At its core, there are three main promises:
- An improvement of care quality
- The reduction of unwarranted and unproven treatment variation
- The “assurance” that limited resources are spent on interventions proven to be effective.
For the complex, fragmented and resource-intensive U.S. health care system, EBM is not merely an aspiration. It is an economic and moral imperative for achieving better outcomes and controlling runaway costs.
CHANGE IS INEVITABLE (and Necessary)
Despite its foundational importance, the practical application of traditional EBM in the U.S. faces persistent, systemic barriers. The sheer scale and speed of modern medical research have created an intractable problem of information overload, while high administrative burdens, heavy patient volume and structural payment models often deny clinicians the necessary time and resources to engage in rigorous evidence appraisal. Put simply, how are we as providers supposed to keep up with the mass quantity of material present in the literature? In our eye care industry alone, there are approximately 200,000 pages written in English every year. If we as professionals read and process at the standard rate of fifteen pages per hour, it will only take us a little over 13,000 hours or 1,666 business days a year to stay current—this year!
The Role of AI
As we look to the future, these statistics may become less overwhelming as artificial intelligence (AI) and machine learning (ML) become integrated into our “science based clinical process.” AI represents a transformative force capable of processing data and generating insights at a speed and scale previously unimaginable, and greater than our individual gray box computers (our brains). It will no longer be just an auxiliary tool; it will augment our knowledge and is poised to become an active participant in providers’ clinical decision-making process for diagnosis, treatment planning and research synthesis. AI will not only augment us across multiple parts of every patient’s health care journey, but it has the capacity to completely shift how we make decisions – and the decision that we make.
AI is poised to solve the systemic implementation failures of EBM within the resource-constrained U.S. health care system—by overcoming the “time and knowledge” barriers. Here comes the “but” though. There are also concerns that AI trained on the wrong models or removing “human in the loop” protocols may exacerbate existing health disparities if implemented without proactive policy and rigorous regulatory alignment. To understand the future of EBM in the U.S., we must first examine the chronic systemic issues AI promises to resolve. Only then can we analyze its transformative potential across the clinical spectrum or measure its likely systemic impact on the health care model.
EBM’s Achilles’ Heel: Systemic Challenges in the U.S.
Integration and implementation of AI into the U.S. system is consistently undermined by three structural challenges.
The Information Overload Crisis
As mentioned, the most pressing challenge is the exponential growth of medical knowledge. The volume of new medical literature, including journal articles, clinical trials and systematic reviews, is now so vast that the available medical knowledge is estimated to be doubling every few months. This information overload creates an insurmountable obstacle for the individual clinician. Many times, providers revert to what they recall from their education, maybe years to decades ago, combined with what their experience has taught them. The inevitable result is the persistent medical knowledge lag: a significant delay between the moment a piece of research evidence is validated and the moment it is reliably and consistently applied in clinical practice nationwide. This lag translates directly into avoidable clinical variation and suboptimal patient outcomes that could be overcome if we have the best validated knowledge at our fingertips on demand or provided proactively.
Resource Constraints
In the American health care environment, the primary barrier to robust EBM implementation is often lack of time. Staffing shortages, particularly in primary care, couple with notoriously high administrative burdens related to charting, billing and regulatory compliance. These factors collectively prevent clinicians from dedicating sufficient time to evidence-based activities, such as literature searches, attending educational sessions on new guidelines or developing and implementing practice change protocols. Add to that fact that there are inherent cognitive and emotional biases present in all of the “evidence-based” activities. Sometimes the “evidence” may not be as “pure” as it seems or is touted. Furthermore, the pervasive fee-for-service payment model in the U.S. structurally incentivizes volume over value. Clinicians are rewarded for the quantity of services provided, not necessarily for the thoughtful, time-intensive process required to fully apply EBM principles, which integrate best evidence with a nuanced understanding of patient values.
Data Silos and Clinical Decision-Making Inconsistencies
Effective EBM requires a feedback loop: evidence must be applied, and the outcomes must be measured to refine practice. There are a few challenges to implementation of this feedback loop. First, the U.S. health care landscape is characterized by fragmented technology, particularly in electronic health records (EHRs). Data silos—where information is locked within specific institutional or vendor systems— makes it immensely difficult to aggregate standardized, timely and useful clinical data across a single patient’s health care journey, much less across various phenotypes or genotypes on a global scale. Compounding this technical challenge is a cultural and organizational resistance to change. Institutional norms or reliance on traditional practices can often outweigh the adoption of new, evidence-supported protocols, leading to inconsistent application of best evidence across the country. O
AI as the Engine for EBM 2.0: Augmenting Clinical Practice
AI will need to offer providers both a compelling and intuitive solution to these systemic failures by effectively giving clinicians a dedicated research assistant, trusted data analyst and an adaptable organizational optimizer. Let’s break these three necessary components down.
Accelerated Evidence Synthesis
Large language models (LLMs) and specialized AI tools represent the most immediate solution to the information overload crisis. These applications can instantly summarize and critically filter vast quantities of unstructured data, including research papers, clinical guidelines and patient records. By providing clinicians with a concise, distilled and ranked summary of the most relevant and highest-quality evidence in seconds, AI has the capacity to dramatically reduce the cognitive load and time constraints barrier. This capability moves us closer to real-time, or EBM 2.0, where evidence is applied concurrent with the clinical question, rather than lagging days or decades behind.
Personalized Precision Diagnosis and Treatment
The complexity of modern diseases demands a level of data integration that exceeds inherent human capacity. AI-powered algorithms excel at analyzing complex, multimodal data, including genetic sequences (genomics), diagnostic imaging and longitudinal patient history in the EHR. If these systems have access to cross platform personal data, AI’s analytic power will allow us to move beyond general guidelines to provide personalized treatment recommendations. By correlating objective “best evidence” with an individual patient’s unique biological profile and historical context, AI will be able to effectively and rigorously integrate the often-neglected “patient values” component of EBM with objective science.
Streamlining Clinical Workflows
Perhaps the most universally welcomed application of AI is its ability to directly attack the problem of administrative burden. By automating tasks such as generating preliminary notes from transcribed patient-physician dialogue, filling out forms, handling initial billing codes and assisting with triage, AI acts as an artificial co-pilot. This streamlining of clinical workflows and documentation will free up a significant portion of a clinician’s day that is currently spent on non-clinical, administrative duties. The resulting time dividend can then be reinvested into direct patient interaction, critical thinking and the very EBM activities that current system demands prevent. By reducing physician burnout and increasing available cognitive bandwidth, AI supports both the implementation of EBM and the overall sustainability of the U.S. health care workforce.
Systemic Impact on the US Health Care Model
The integration of AI into EBM promises profound systemic effects. It offers a potential pathway to achieving three major goals of better health, better care and lower costs.
Improving Quality and Patient Outcomes
The standardization of care is a direct benefit of AI integration. By embedding the most current, rigorously appraised EBM directly into AI-driven clinical decision support (CDS) tools either within the EHR or into yet undiscovered knowledge platforms, health systems can ensure that every patient, regardless of their location or the time of day, receives a level of care aligned with consensus best practices. Furthermore, AI’s superiority in pattern recognition can significantly reduce diagnostic and medical errors. Tools capable of interpreting diagnostic images or flagging potentially dangerous drug interactions act as an essential safety net, improving reliability and reducing avoidable harm.
Cost Reduction and Efficiency
In a system plagued by high operational costs, AI offers tangible pathways to efficiency. By providing predictive analytics regarding patient compliance, treatment response and foresight of certain resource requirements, AI can optimize resource allocation, ensuring that staffing and supplies are managed proactively rather than reactively. The ability to accelerate drug discovery and development has the potential to dramatically reduce the staggering R&D costs that contribute to high drug prices in the U.S. While the initial investment in AI infrastructure is high, the long-term potential for reduced waste, fewer readmissions, and lower pharmaceutical costs is immense.
Addressing Health Access Disparities
The impact of AI on health equity presents a crucial double-edged sword. On one hand, AI holds positive potential for addressing access barriers. A single, centralized AI diagnostic system can support under-resourced or remote facilities, effectively distributing the expertise of a world-class specialist to areas where human expertise is scarce. This democratization of high-quality, evidence-based expertise can uplift the standard of care in rural and underserved communities.
However, the negative risk of exacerbating existing health inequities is significant. If AI models are trained predominantly on historical data sets that underrepresent marginalized populations, the resulting algorithms will be inherently biased. This can lead to disparate impacts, such as algorithms assigning lower risk scores to Black patients with similar conditions compared to White patients, simply because less money has historically been spent on their care. Such biased tools would solidify and worsen racial and socioeconomic disparities in care, actively undermining the equity goals of the U.S. health care system.
The Regulatory and Ethical Chasm: Guardrails for the Future
The promises of AI-driven EBM can only be realized if they are anchored by a universally accepted robust and proactive regulatory framework designed to mitigate significant ethical risks.
Bias, Fairness and Data Integrity
The principal ethical challenge is ensuring fairness and data integrity. The performance of any AI system is only as good as the data it consumes.. A critical policy need is the mandatory development and use of diverse training data sets that are rigorously validated for different demographic groups, coupled with regular, standardized bias audits throughout the AI lifecycle. This requires a commitment to proactive intervention to address the perpetuation of health inequities.
Transparency and the “Black Box” Problem
Many powerful deep-learning AI systems operate as “black boxes.” This means that the complexity of their internal function makes it difficult for clinicians, regulators and even the developers to fully explain why a particular decision or recommendation was made. This opacity creates a critical challenge for EBM, which relies on being able to critically appraise and validate the rationale behind a clinical action.
Another policy need is to establish standards for explainable AI (XAI). This explanation mandate should be proportionate to the risk level of the application—higher transparency should be required for life-critical diagnostic or autonomous surgical AI than for administrative tools. Clinicians must be able to understand the basis of the recommendation to integrate it effectively with their clinical judgment.
Accountability and Liability
Finally, the question of accountability poses a complex legal challenge. When an AI-driven decision—whether it be a misdiagnosis or a flawed treatment plan—results in patient harm, current legal frameworks struggle to determine liability. Is the fault with the prescribing clinician (whose judgment the AI influenced), the business (which implemented the technology) or the AI developer (who built the faulty algorithm)? The current standard of care and existing malpractice law are ill-equipped to handle this ambiguity. This will likely require a transparent redefinition of standards of care and the clarification of legal frameworks for AI use in malpractice law, ensuring that accountability for AI errors is clearly defined and defensible. In many cases, a move towards strict liability for developers of autonomous, high-risk systems may be necessary.
Patient Trust and Informed Consent
Finally, and most importantly, the integrity of the patient-physician relationship must be preserved. The use of AI, particularly in sensitive areas like mental health diagnostics or end-of-life care planning, requires that patients fully understand the extent of AI involvement in their care. Informed consent must evolve to encompass the disclosure of AI usage, allowing patients the right to opt-out if they are uncomfortable. The fragmented and unstructured massive datasets required to train effective AI models also raise significant patient privacy concerns, necessitating robust enforcement and potentially updated interpretations of existing data privacy laws like HIPAA, ensuring that patient information is securely protected as it is aggregated and processed.
Conclusion
AI integration is not an optional add-on. It is the inevitable next phase of evidence-based medicine. This process also counters chronic systemic failures in U.S. health care. It addresses information overload and resource constraints.
An integrated AI-EBM approach is arguably the only sustainable path. It delivers high-quality, equitable cost-effective U.S. health care. The challenge now is not technological, but political and ethical—and emotional.
Author
