Unpaid AI work: How it limits care for your patients

The deployment of artificial intelligence in primary care and specialist settings presents a clear opportunity to enhance diagnostic accuracy, streamline administrative tasks, and personalise treatment plans. But the economic reality for clinicians is that time spent evaluating AI outputs, integrating them into patient care, and managing the associated data infrastructure is currently uncompensated. This disconnect between technological advancement and financial recognition creates a significant barrier to widespread adoption, even for tools that demonstrably improve efficiency or patient safety.

Consider a GP using an AI-powered diagnostic assistant for early detection of diabetic retinopathy. The AI might analyse retinal scans faster and with greater consistency than a human eye, flagging subtle changes that could otherwise be missed. The clinician's role shifts from primary interpretation to validation and patient communication, a process that still demands time and clinical judgment. This augmented workflow, while beneficial for the patient, does not fit neatly into existing billing codes for a standard eye exam or consultation.

The Challenge of Value in AI-Augmented Care

The fundamental problem lies in defining the 'value' of AI in a way that translates into a billable service. Traditional fee-for-service models pay for discrete actions: a consultation, a procedure, a lab test. AI, however, often acts as an assistive technology, augmenting human capability rather than replacing it entirely. How does one bill for the 'insight' provided by an algorithm, or the time saved by automated documentation, when the human clinician remains ultimately responsible for the patient's care?

Some AI applications, such as automated medical scribes, promise to reduce the administrative burden on clinicians, freeing up time for direct patient interaction. This efficiency gain is valuable, but it does not generate a new billable event. Instead, it potentially reduces the time spent on existing, billable tasks. While this improves clinician satisfaction and reduces burnout, it does not directly increase revenue under a fee-for-service model, making the investment in such tools a purely overhead cost.

Value-based care models, which tie reimbursement to patient outcomes and quality metrics, might seem a more natural fit for AI. If an AI tool demonstrably improves a quality metric, such as reducing hospital readmissions or improving adherence to preventative screenings, then the health system or payer could theoretically incentivise its use. But establishing a direct causal link between a specific AI tool and a population-level outcome, while controlling for all other variables, is a complex epidemiological and statistical challenge. The data infrastructure required to track and attribute these outcomes accurately is often lacking, particularly in smaller practices.

The development and deployment of AI tools often involve significant upfront costs, from software licenses to integration with existing electronic health records. These costs must be recouped, but without clear reimbursement pathways, individual practices or smaller health systems may find the investment prohibitive. This creates a disparity, where larger, better-resourced institutions are more likely to adopt AI, potentially widening the gap in care quality and efficiency across the healthcare system. The broader implications for health policy and infrastructure are substantial, as discussed in our previous coverage on infrastructure flaws.

Navigating New Reimbursement Pathways

Creating viable reimbursement pathways for AI tools will require a multi-pronged approach, involving both new coding structures and innovative payment models. One immediate need is the development of specific Current Procedural Terminology (CPT) codes or their European equivalents that recognise AI-augmented services. These codes could account for the clinician's time in reviewing AI outputs, validating diagnoses, or using AI for complex treatment planning.

For instance, a new code might be established for 'AI-assisted diagnostic review' or 'AI-guided treatment optimisation'. These codes would need to be carefully defined to prevent misuse and ensure that they genuinely reflect a distinct clinical service. The challenge lies in standardising these definitions across diverse AI applications and clinical specialties, from AI in neurology to tools for chronic disease management.

Another approach involves bundling AI services into existing care pathways, with an uplift in reimbursement for the entire bundle if AI is used and specific quality metrics are met. This would align with value-based care principles, incentivising the use of AI where it can demonstrate tangible improvements in patient outcomes or cost-effectiveness. But this requires robust data collection and attribution mechanisms, which are still nascent in many healthcare settings.

Payers could also consider direct payments to practices for the adoption and integration of certified AI tools, perhaps through grants or performance-based incentives. This would help offset the initial investment costs and encourage early adoption. Such models would need to be carefully structured to ensure that the AI tools being incentivised are clinically validated and provide genuine benefit, rather than simply being novel technologies. The regulatory landscape, which refers to the guidelines around AI in healthcare, is still evolving, and clear guidelines on validation and efficacy are essential.

The conversation around AI reimbursement also touches on the broader issue of physician workload and the value of non-face-to-face clinical activities. If AI can genuinely reduce documentation time, for example, then perhaps existing consultation codes should be re-evaluated to reflect the increased cognitive load of patient interaction. This would be a significant shift, but one that acknowledges the changing nature of clinical work in an AI-enhanced environment. The ethical and legal considerations of AI scribes, for example, are already being debated.

The Path Forward for Clinicians and Health Systems

For individual clinicians and practices, the immediate path involves careful evaluation of AI tools for their genuine clinical utility and return on investment, even in the absence of direct reimbursement. Prioritising tools that address significant pain points, such as administrative burden or diagnostic uncertainty, can still yield benefits in terms of efficiency and job satisfaction. Understanding the nuances of these tools is critical, much like mastering the Oxford Handbook of Clinical Medicine for daily practice.

Health systems and professional organisations have a crucial role to play in advocating for policy changes, with the actual stake being the widespread adoption and equitable access to beneficial AI technologies. This includes engaging with payers, regulatory bodies, and government agencies to develop new reimbursement frameworks that are fit for purpose in the age of AI. Pilot programs and real-world evidence generation will be vital in demonstrating the economic and clinical value of AI-augmented care, providing the robust data (n=1200, 95% CI) needed to justify new payment models.

The long-term success of AI integration in healthcare hinges not just on technological prowess, but on the ability of the healthcare system to adapt its financial and operational structures. Without a clear and equitable way to compensate clinicians for their use of AI, the promise of these tools will remain largely unfulfilled, confined to early adopters or well-funded institutions. The challenge is to move beyond simply acknowledging AI's potential and to actively build the economic scaffolding required for its widespread, beneficial application.